CN118038283A

CN118038283A - Method and equipment for detecting hidden diseases of asphalt pavement

Info

Publication number: CN118038283A
Application number: CN202410449157.6A
Authority: CN
Inventors: 李法滨; 刘万军; 梁栋; 沈兆坤; 袁超; 廖连山; 蔡文俊; 张倪; 杜铁军; 杨莉; 张伟
Original assignee: Guizhou Qiantong Engineering Technology Co ltd
Current assignee: Guizhou Qiantong Engineering Technology Co ltd
Priority date: 2024-04-15
Filing date: 2024-04-15
Publication date: 2024-05-14
Anticipated expiration: 2044-04-15
Also published as: CN118038283B

Abstract

The application provides a method and equipment for detecting blind diseases of an asphalt pavement, which can be used for pre-determining candidate confidence levels of multiple classes of set pavement disease classifications corresponding to an asphalt pavement radar imaging image based on the set pavement disease classifications corresponding to confirmed past reference detection information, then carrying out disease detection based on the image and the corresponding candidate confidence levels, and determining estimated confidence levels of the multiple classes of set pavement disease classifications. Because the probability that the corresponding set pavement disease classification is consistent with the set pavement disease classification corresponding to the confirmed past reference detection information is high, in addition to the image, the set pavement disease classification corresponding to the confirmed past reference detection information is also introduced for auxiliary analysis, the set pavement disease classification corresponding to the confirmed past reference detection information is used as a target, and the asphalt pavement radar imaging image is analyzed by combining the asphalt pavement radar imaging image, so that the accuracy of disease detection on the asphalt pavement radar imaging image is improved.

Description

Method and equipment for detecting hidden diseases of asphalt pavement

Technical Field

The application relates to the technical fields of image processing and artificial intelligence, in particular to a method and equipment for detecting hidden diseases of an asphalt pavement.

Background

With the rapid development of traffic infrastructure construction, asphalt pavement is widely used due to its excellent running performance and low maintenance cost. However, with the increase of the service time, various hidden diseases such as cracks, pits and the like, which seriously affect the service life and the driving safety of the road, inevitably occur on the asphalt pavement. Therefore, it is particularly important to accurately and timely detect the hidden diseases on the asphalt pavement.

At present, the traditional method for detecting the hidden diseases of the asphalt pavement mainly depends on manual inspection and visual judgment, but the method is low in efficiency, limited by experience and skill level of inspection personnel, and difficult to ensure the accuracy and consistency of detection. In recent years, along with development of radar technology, radar imaging technology is introduced into the detection of the hidden diseases of the asphalt pavement, and the radar imaging technology can penetrate through the surface layer of the pavement to acquire deeper internal structure information, so that the hidden diseases can be found in advance, and a new possibility is provided for the detection of the diseases.

However, existing radar imaging techniques still present challenges in asphalt pavement cryptogamic disease detection. On the one hand, radar imaging images often contain a lot of noise and interference information due to the complexity of radar signals and the non-uniformity of asphalt pavement, which makes disease identification difficult. On the other hand, the asphalt pavement has various hidden diseases and various forms, and the different diseases have larger difference in the radar imaging images, so that the difficulty of disease detection is further increased. In other words, the accuracy of the prior art for detecting the hidden diseases of the asphalt pavement needs to be improved.

Disclosure of Invention

Therefore, the embodiment of the application at least provides a method and equipment for detecting the hidden diseases of the asphalt pavement. The technical scheme of the embodiment of the application is realized as follows:

In one aspect, the embodiment of the application provides a method for detecting a hidden disease of an asphalt pavement, which comprises the following steps: acquiring an asphalt pavement radar imaging image to be subjected to disease detection and a plurality of pieces of reference detection information corresponding to the asphalt pavement radar imaging image, wherein the plurality of pieces of reference detection information comprise confirmed past reference detection information corresponding to the asphalt pavement radar imaging image; according to the set pavement disease classification corresponding to the multiple reference detection information, determining candidate pavement disease type information of the asphalt pavement radar imaging image; the candidate pavement disease type information comprises candidate confidence degrees of the asphalt pavement radar imaging images corresponding to multiple types of set pavement disease classifications; acquiring an integrated characterization vector of the asphalt pavement radar imaging image according to the asphalt pavement radar imaging image and the candidate pavement disease type information; performing disease detection on the asphalt pavement radar imaging image according to the integrated characterization vector to obtain a predicted disease detection result of the asphalt pavement radar imaging image; the estimated disease detection result comprises estimated confidence degrees of the asphalt pavement radar imaging image corresponding to the plurality of set pavement disease classifications, and the estimated disease detection result is used for determining the set pavement disease classifications corresponding to the asphalt pavement radar imaging image.

In some embodiments, the determining the candidate pavement disease type information of the asphalt pavement radar imaging image according to the set pavement disease classifications corresponding to the plurality of reference detection information includes: acquiring the number Q1 of each set pavement disease classification, wherein the number Q1 is the number of past reference detection information which belongs to the set pavement disease classification and is confirmed in the plurality of reference detection information; and taking a first quotient result between the number Q1 of the set road surface disease classifications of each type and the total number of the plurality of reference detection information as a candidate confidence coefficient corresponding to the set road surface disease classifications of each type.

In some embodiments, the plurality of reference detection information further includes repudiated past reference detection information corresponding to the asphalt pavement radar imaging image; the determining candidate pavement disease type information of the asphalt pavement radar imaging image according to the set pavement disease classification corresponding to the plurality of reference detection information comprises the following steps: acquiring the number Q2 of each set pavement disease classification, wherein the number Q2 is the number of past reference detection information belonging to the set pavement disease classification in the plurality of reference detection information; obtaining a second quotient result between the number Q2 of each set pavement disease classification and the total number of the plurality of reference detection information; acquiring the number Q3 of each set pavement disease classification, wherein the number Q3 is the number of past reference detection information which belongs to the set pavement disease classification and is confirmed in the plurality of reference detection information; obtaining a third quotient result between the number Q3 of each set pavement disease classification and the total number of confirmed past reference detection information; and respectively carrying out weight average on the second and third provider results of each class of set pavement disease classification to obtain candidate confidence degrees corresponding to each class of set pavement disease classification.

In some embodiments, the performing disease detection on the asphalt pavement radar imaging image according to the integrated characterization vector to obtain a predicted disease detection result of the asphalt pavement radar imaging image includes: extracting projection characterization vectors of a plurality of channels according to the integration characterization vectors; according to the integrated characterization vector, obtaining an influence coefficient set of each type of set pavement disease classification, wherein the influence coefficient set comprises influence coefficients of the channels, and the influence coefficients represent the contribution degree of characterization information of an image on the channel to whether the image is the set pavement disease classification; respectively carrying out weight average on projection characterization vectors of the channels according to a plurality of influence coefficients in the influence coefficient set of each type of set pavement disease classification to obtain a type characterization vector of each type of set pavement disease classification; and respectively carrying out disease detection on the asphalt pavement radar imaging image according to the type characterization vector of each type of set pavement disease classification to obtain the estimated confidence corresponding to each type of set pavement disease classification.

In some embodiments, after the disease detection is performed on the asphalt pavement radar imaging image according to the integrated characterization vector to obtain the estimated disease detection result of the asphalt pavement radar imaging image, the method further includes: taking the set pavement disease classification with the maximum estimated confidence as the set pavement disease classification corresponding to the asphalt pavement radar imaging image; determining a correlation coefficient between the asphalt pavement radar imaging image and each piece of undetermined reference detection information according to the set pavement disease classification corresponding to the asphalt pavement radar imaging image and the set pavement disease classification corresponding to the plurality of undetermined reference detection information corresponding to the asphalt pavement radar imaging image; taking undetermined reference detection information with the association coefficient meeting the association coefficient requirement as target detection information determined according to the asphalt pavement radar imaging image; before determining the candidate pavement disease type information of the asphalt pavement radar imaging image according to the set pavement disease classifications corresponding to the plurality of reference detection information, the method further comprises: acquiring a plurality of road surface disease classifications corresponding to pre-stored undetermined reference detection information; acquiring the number Q4 corresponding to each road surface disease classification, wherein the number Q4 is the number of the pre-stored undetermined reference detection information belonging to the road surface disease classification; and taking a plurality of road surface defect classifications, the number Q4 of which meets a number threshold value, as the plurality of set road surface defect classifications.

In some embodiments, the radar imaging image detection network includes a characterization vector extraction component and a disease detection component, the obtaining an integrated characterization vector of the asphalt pavement radar imaging image from the asphalt pavement radar imaging image and the candidate pavement disease type information includes: the integrated characterization vector is obtained according to the asphalt pavement radar imaging image and the candidate pavement disease type information through the characterization vector extraction component; the disease detection is carried out on the asphalt pavement radar imaging image according to the integrated characterization vector to obtain a predicted disease detection result of the asphalt pavement radar imaging image, and the method comprises the following steps: and performing disease detection on the asphalt pavement radar imaging image according to the integrated characterization vector by the disease detection assembly to obtain the estimated disease detection result.

In some embodiments, the radar imaging image detection network is commissioned by: acquiring a first asphalt pavement radar imaging image sample and a plurality of sample reference detection information corresponding to the first asphalt pavement radar imaging image sample, wherein the sample reference detection information comprises confirmed past reference detection information corresponding to the first asphalt pavement radar imaging image sample; determining first candidate pavement disease type information of the first asphalt pavement radar imaging image sample according to set pavement disease classifications corresponding to the multiple sample reference detection information, wherein the first candidate pavement disease type information comprises sample candidate confidence coefficients of the first asphalt pavement radar imaging image sample corresponding to the multiple set pavement disease classifications; obtaining a sample integration characterization vector of the first asphalt pavement radar imaging image sample according to the first asphalt pavement radar imaging image sample and the first candidate pavement disease type information through the characterization vector extraction component; performing disease detection on the first asphalt pavement radar imaging image sample according to the sample integration characterization vector through the disease detection assembly to obtain a first estimated disease detection result, wherein the first estimated disease detection result comprises sample estimated confidence degrees of the first asphalt pavement radar imaging image sample corresponding to the plurality of set pavement disease classifications; according to a first priori disease detection result and the first estimated disease detection result of the first asphalt pavement radar imaging image sample, the radar imaging image detection network is debugged, so that errors between the first estimated disease detection result and the first priori disease detection result obtained by the radar imaging image detection network after debugging are reduced, and the first priori disease detection result comprises actual confidence that the first asphalt pavement radar imaging image sample corresponds to the classification of the various set pavement diseases.

In some embodiments, the token vector extraction component includes a first extraction unit, a second extraction unit, and a token information interaction unit, and the obtaining, by the token vector extraction component, a sample integration token vector of the first asphalt pavement radar imaging image sample according to the first asphalt pavement radar imaging image sample and the first candidate pavement disease type information includes: extracting, by the first extracting unit, a sample image characterization vector of the first asphalt pavement radar imaging image sample; extracting sample disease information characterization vectors of the first candidate pavement disease type information through the second extraction unit; the sample image characterization vector and the sample disease information characterization vector are subjected to interactive integration through the characterization information interaction unit to obtain the sample integration characterization vector; the disease detection assembly comprises a characterization vector projection unit corresponding to a plurality of channels, a weight adjustment unit for setting road surface disease classification of each class and a disease identification unit; the disease detection module is used for performing disease detection on the first asphalt pavement radar imaging image sample according to the sample integration characterization vector to obtain a first estimated disease detection result, and the method comprises the following steps: respectively integrating the characterization vectors according to the samples by a plurality of characterization vector projection units, and extracting sample projection characterization vectors of the channels; the method comprises the steps that a weight adjusting unit of each type of set pavement disease classification is used for integrating a characterization vector according to the sample, and an influence coefficient set of each type of set pavement disease classification is obtained, wherein the influence coefficient set comprises influence coefficients of a plurality of channels, and the influence coefficients represent the contribution degree of characterization information of an image on the channels to whether the image is the set pavement disease classification or not; respectively carrying out weight average on sample projection characterization vectors of the channels according to a plurality of influence coefficients in an influence coefficient set of each type of set pavement disease classification by a weight adjusting unit of each type of set pavement disease classification to obtain sample type characterization vectors of each type of set pavement disease classification; and respectively carrying out disease detection on the first asphalt pavement radar imaging image samples according to the sample type characterization vectors of each type of set pavement disease classification by the disease identification unit of each type of set pavement disease classification to obtain the sample prediction confidence corresponding to each type of set pavement disease classification.

In some embodiments, after the debugging the radar imaging image detection network according to the first priori disease detection result and the first estimated disease detection result of the first asphalt pavement radar imaging image sample, the method further includes: acquiring a second asphalt pavement radar imaging image sample, and taking set pavement disease classification information as second candidate pavement disease type information of the second asphalt pavement radar imaging image sample, wherein the second candidate pavement disease type information comprises sample candidate confidence coefficients of the second asphalt pavement radar imaging image sample corresponding to the plurality of set pavement disease classifications; obtaining a sample integration characterization vector of the second asphalt pavement radar imaging image sample according to the second asphalt pavement radar imaging image sample and the second candidate pavement disease type information through the characterization vector extraction component; performing disease detection on the second asphalt pavement radar imaging image sample according to the sample integration characterization vector through the disease detection assembly to obtain a second estimated disease detection result, wherein the second estimated disease detection result comprises sample estimated confidence degrees of the second asphalt pavement radar imaging image sample corresponding to the plurality of set pavement disease classifications; according to a second priori disease detection result and the second estimated disease detection result of the second asphalt pavement radar imaging image sample, the radar imaging image detection network is debugged, so that errors between the second estimated disease detection result obtained by the radar imaging image detection network after debugging and the second priori disease detection result are reduced, and the second priori disease detection result comprises actual confidence that the second asphalt pavement radar imaging image sample corresponds to the classification of the various set pavement diseases.

In some embodiments, the debugging the radar imaging image detection network according to the first priori disease detection result and the first estimated disease detection result of the first asphalt pavement radar imaging image sample, so that an error between the first estimated disease detection result and the first priori disease detection result obtained by the radar imaging image detection network after the debugging is reduced, includes: according to cost influence coefficients corresponding to each set pavement disease classification, carrying out weight average on cross entropy between the estimated confidence coefficient and the actual confidence coefficient of the sample of each set pavement disease classification to obtain a cost parameter, wherein the degree of commonality between the estimated confidence coefficient and the actual confidence coefficient of the sample is inversely related to the cross entropy; and according to the cost parameter, debugging the radar imaging image detection network, and adjusting the cost influence coefficient, so that the cost parameter obtained according to the radar imaging image detection network after the debugging and the adjusted cost influence coefficient is reduced.

In a second aspect, the application provides a computer device comprising a memory and a processor, the memory storing a computer program executable on the processor, the processor implementing the steps of the method described above when the program is executed.

The technical effects of the application at least comprise: according to the method and the device for detecting the blind diseases of the asphalt pavement, provided by the application, the imaging image of the radar of the asphalt pavement is related to the confirmed past reference detection information, so that the imaging image of the radar of the asphalt pavement can be guided by the past information to comprehensively evaluate the current state. Then, based on the confirmed set pavement disease classification corresponding to the past reference detection information, the candidate confidence coefficient of the asphalt pavement radar imaging image corresponding to the multi-class set pavement disease classification can be predetermined, then, based on the asphalt pavement radar imaging image and the corresponding candidate confidence coefficient, disease detection is carried out, and the estimated confidence coefficient of the asphalt pavement radar imaging image corresponding to the multi-class set pavement disease classification is determined. Because the probability that the set pavement disease classification corresponding to the asphalt pavement radar imaging image is consistent with the set pavement disease classification corresponding to the confirmed past reference detection information is higher, the set pavement disease classification corresponding to the confirmed past reference detection information is also introduced for auxiliary analysis besides the asphalt pavement radar imaging image, the set pavement disease classification corresponding to the confirmed past reference detection information is used as a standard, and the asphalt pavement radar imaging image is analyzed by combining the asphalt pavement radar imaging image, so that the accuracy of disease detection on the asphalt pavement radar imaging image is improved.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and together with the description, serve to explain the principles of the application.

Fig. 1 is a schematic implementation flow chart of a method for detecting a hidden disease of an asphalt pavement according to an embodiment of the present application.

Fig. 2 is a schematic diagram of a hardware entity of a computer device according to an embodiment of the present application.

Detailed Description

The technical solution of the present application will be further elaborated with reference to the accompanying drawings and examples, which should not be construed as limiting the application, but all other embodiments which can be obtained by one skilled in the art without making inventive efforts are within the scope of protection of the present application.

In the following description, reference is made to "some embodiments" which describe a subset of all possible embodiments, but it is to be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments and can be combined with one another without conflict. The term "first/second/third" is merely to distinguish similar objects and does not represent a particular ordering of objects, it being understood that the "first/second/third" may be interchanged with a particular order or precedence, as allowed, to enable embodiments of the application described herein to be implemented in other than those illustrated or described herein.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing the application only and is not intended to be limiting of the application.

The embodiment of the application provides a method for detecting a hidden disease of an asphalt pavement, which can be executed by a processor of computer equipment. The computer device may refer to a server, a notebook computer, a tablet computer, a desktop computer, or other devices with data processing capability.

Fig. 1 is a schematic implementation flow chart of a method for detecting a hidden disease of an asphalt pavement, which is provided by an embodiment of the application, and as shown in fig. 1, the method comprises the following steps:

Step S100: the method comprises the steps of obtaining an asphalt pavement radar imaging image to be subjected to disease detection and a plurality of pieces of reference detection information corresponding to the asphalt pavement radar imaging image, wherein the plurality of pieces of reference detection information comprise confirmed past reference detection information corresponding to the asphalt pavement radar imaging image.

In step S100, the asphalt pavement radar imaging image is a result obtained after the computer device detects the asphalt pavement of the target road section by the ground penetrating radar. The ground penetrating radar can penetrate through the asphalt layer, capture structural information in the pavement, and display the structural information in the form of an image so as to help identify hidden diseases. The images contain rich pavement condition information, such as the thickness of an asphalt layer, whether cavities, cracks and other potential diseases exist in the images. The reference detection information refers to past disease detection results which are associated with these radar imaging images and have been confirmed. These past results are obtained by historical detection activities, which reflect the disease condition of the target road segment at different points in time and/or at different locations. Such past information is critical to computer devices because they provide valuable clues about the occurrence, development, and variation of the target road segment diseases.

For example, assume that a computer device is performing asphalt pavement disease detection for a arterial road of a city. In step S100, first, a radar imaging image of the arterial road is acquired by a ground penetrating radar. These images may show abnormal reflection patterns in certain areas, suggesting possible internal road surface imperfections.

Meanwhile, the computer equipment can also acquire the past disease detection record of the main road, namely the reference detection information. These records may include all occurrences of disease that occurred in the road segment over the last few years, such as cracks, potholes, etc. in the road surface. This past disease information is used to compare and analyze with current radar imaging images to help computer equipment more accurately identify and locate potential disease areas.

Step S200: according to the set pavement disease classification corresponding to the multiple reference detection information, determining candidate pavement disease type information of the asphalt pavement radar imaging image; the candidate pavement disease type information comprises candidate confidence degrees of the asphalt pavement radar imaging images corresponding to the multi-type set pavement disease classifications.

In step S200, the computer device determines the kind of road surface disease that may exist in the current asphalt pavement radar imaging image based on the plurality of reference detection information previously acquired, which correspond to the known road surface disease classifications. In particular, the computer device first analyzes the reference detection information, which typically includes the type and location of the disease that was historically identified when detecting the disease for the same or similar road segments. For example, if a past detection record shows that two diseases, namely a crack and a pit, often occur on a certain road section, the two diseases are regarded as important attention objects of the current detection.

Then, the computer device performs preliminary screening and judgment on the current radar imaging image of the asphalt pavement according to the known disease classifications, and determines which areas in the image are likely to belong to the disease types. This process typically involves the application of image processing techniques and machine learning algorithms. For example, the computer device may use an edge detection algorithm to identify possible cracks in the image, or a texture analysis algorithm to detect pothole areas.

After determining candidate pavement defect areas, the computer device also calculates a candidate confidence level for each area. This confidence reflects the amount of likelihood that the computer device believes that there is some disease in the area. The confidence level is typically calculated based on a variety of factors including the degree of matching of the image features, statistical distribution of historical disease data, and the like. For example, if the image features of a region are highly similar to typical features of a crack, and the region also frequently shows cracks in historical data, the computer device assigns a higher candidate confidence to the region.

As an embodiment, step S200, the determining candidate pavement disease type information of the asphalt pavement radar imaging image according to the set pavement disease classifications corresponding to the multiple reference detection information may specifically include:

step S201: acquiring the number Q1 of each set pavement disease classification, wherein the number Q1 is the number of past reference detection information which belongs to the set pavement disease classification and is confirmed in the plurality of reference detection information;

Step S202: and taking a first quotient result between the number Q1 of the set road surface disease classifications of each type and the total number of the plurality of reference detection information as a candidate confidence coefficient corresponding to the set road surface disease classifications of each type.

In step S201, the core task of the computer apparatus is to acquire the number Q1 of classification of each set road surface disease. This number Q1 actually represents the number of past detection records confirmed to belong to a certain specific road surface disease type among the plurality of reference detection information collected previously.

For example, a city has many asphalt roads that have historically been monitored by computer equipment, accumulating a large number of disease detection records. These records contain various disease types such as road cracks, potholes, water damage, etc. Now, the computer device wants to know what the number of times each disease type is confirmed in these histories.

First, the computer device screens out all past reference detection information about the cracks, and calculates the number of crack defects confirmed in the information to obtain the number Q1 of the cracks. For example, if there are 8 records that clearly confirm the existence of a crack among the past 10 crack-related detection records, the number Q1 of cracks is 8. Next, the same process is performed for other types of diseases such as potholes, water damage, and the like. For example, if 6 records of past detection information of pits are confirmed as true pit diseases, the number Q1 of pits is 6. Similarly, if 5 records are identified as water damage diseases in the past information of water damage, the number Q1 of water damage is 5. This process is critical to the subsequent determination of candidate road surface disease type information because it allows the computer device to identify which disease types are more common in the past and which are likely to occur less based on statistical features of historical data. In this way, the computer device can give priority to those disease types that occur at a high frequency and give more care and resource allocation in subsequent detection.

In step S202, the computer device determines a first result, i.e., a ratio, between the number Q1 of each set road surface disease classifications and the total number of all reference detection information as a candidate confidence level corresponding to the type of disease. This candidate confidence is in fact a probability value reflecting the frequency or likelihood of occurrence of a particular type of disease in all known test records.

For example, three types of disease, namely asphalt pavement cracks, potholes and water damage, continue to be used. Assuming that the computer device has completed step S201, the number Q1 of each disease is obtained: the cracks were 3, the pits were 4, and the water damage was 3. Meanwhile, we know that the total reference detection information piece number is 10. The computer device now proceeds to step S202, where it begins to calculate candidate confidence levels for each disease. For a fracture, its candidate confidence is calculated as follows: dividing the number of cracks Q1 (3) by the total number of pieces of reference test information (10) gives 0.3 (or 3%). This means that in all known detection records the probability of crack occurrence is 30%. Similarly, the candidate confidence for a dimple is the number of dimples Q1 (4) divided by the total number of dimples (10), yielding 0.4 (or 4%). The confidence of water damage candidates is then the number of water damage Q1 (3) divided by the total number of bars (10), also 0.3 (or 30%).

These candidate confidence levels are critical to subsequent disease identification and treatment. They not only provide a statistical probability of disease occurrence, but also help computer equipment to prioritize those more common types of disease that may cause greater damage to the pavement. For example, since the confidence of the candidates for the potholes is highest (40%), the computer device may pay more attention to the potential areas of the pothole disease when processing the new radar imaging image, thereby improving the accuracy and efficiency of detection.

In another embodiment, the plurality of reference detection information further includes repudiated past reference detection information corresponding to the asphalt pavement radar imaging image; step S200, determining candidate pavement disease type information of the asphalt pavement radar imaging image according to the set pavement disease classifications corresponding to the plurality of reference detection information, may specifically include:

Step S210: acquiring the number Q2 of each set pavement disease classification, wherein the number Q2 is the number of past reference detection information belonging to the set pavement disease classification in the plurality of reference detection information; obtaining a second quotient result between the number Q2 of each set pavement disease classification and the total number of the plurality of reference detection information;

Step S220: acquiring the number Q3 of each set pavement disease classification, wherein the number Q3 is the number of past reference detection information which belongs to the set pavement disease classification and is confirmed in the plurality of reference detection information; obtaining a third quotient result between the number Q3 of each set pavement disease classification and the total number of confirmed past reference detection information;

Step S230: and respectively carrying out weight average on the second and third provider results of each class of set pavement disease classification to obtain candidate confidence degrees corresponding to each class of set pavement disease classification.

In another embodiment of step S200, the computer device considers not only the confirmed past reference detection information but also the denied past reference detection information, thereby more comprehensively evaluating candidate road surface disease type information of the asphalt road surface radar imaging image.

In step S210, the computer apparatus first obtains the number Q2 of each set road surface disease classification, the number Q2 representing the total number of past reference detection information belonging to a certain specific disease classification among all reference detection information (including confirmed and denied). Then, a second quotient result is calculated between the number Q2 of each type of disease and the total number of all the reference detection information. This second operator result reflects the frequency of occurrence of the disease in all test records (whether confirmed or not).

For example, assuming that there are 10 pieces of reference detection information in which there are 5 pieces of information about the cracks (including confirmed and denied), the number Q2 of cracks is 5, and the second quotient result is 5/10=0.5.

In step S220, the computer device further obtains the number Q3 of each set road surface disease classification, where the number Q3 is specifically indicative of the number belonging to a specific disease classification in the confirmed past reference detection information. Then, a third quotient result between the number Q3 of each type of disease and the total number of confirmed past reference detection information is calculated. This third operator result represents the proportion of occurrence of the disease in the confirmed record.

Continuing with the example above, if, of the 10 pieces of information, 3 records are confirmed as cracks, the number Q3 of cracks is 3, and the third operator results in 3/(assuming that the confirmed total is 6) =0.5 (assuming that a total of 6 records are confirmed as a disease).

In step S230, the computer device performs weight average (i.e. weighted summation) on the two provider results obtained in step S210 and step S220, so as to obtain the candidate confidence corresponding to each set road surface disease classification. The candidate confidence comprehensively considers the frequency of diseases in all detection records and the proportion of diseases in confirmed records, thereby providing a more accurate and comprehensive assessment index.

In the above example, the second quotient result for the crack is 0.5 and the third quotient result is also 0.5 (for example only, the actual values may be different). If the computer device gives equal weight to the two results, then the candidate confidence for the fracture is (0.5+0.5)/2=0.5. This value will be used as a reference basis for subsequent disease identification and treatment.

In this way, the computer device can more accurately evaluate the possibility of different road surface diseases, thereby optimizing resource allocation and improving detection efficiency.

As an embodiment, in step S200, before determining the candidate pavement disease type information of the asphalt pavement radar imaging image according to the set pavement disease classifications corresponding to the plurality of reference detection information, the method may further include:

Step S20a: acquiring a plurality of road surface disease classifications corresponding to pre-stored undetermined reference detection information;

Step S20b: acquiring the number Q4 corresponding to each road surface disease classification, wherein the number Q4 is the number of the pre-stored undetermined reference detection information belonging to the road surface disease classification;

step S20c: and taking a plurality of road surface defect classifications, the number Q4 of which meets a number threshold value, as the plurality of set road surface defect classifications.

The purpose of step S20a is to obtain, from a database of the computer device, a plurality of road surface disease classifications corresponding to the predetermined reference detection information stored in advance. These pending reference test information are data collected before but not yet confirmed in the end, and may contain various types of road surface diseases such as cracks, potholes, water damage, etc.

In performing step S20a, the computer device first accesses a database or other storage medium within it, which stores a large amount of pending reference detection information. Each piece of information is associated with one or more classifications of road surface disease, which are derived based on preliminary observations, sensor readings, or other automated methods.

For example, assume that 1000 pieces of pending reference test information are stored in a database of a computer device, which are collected at different times, places and road conditions. Some of the information may have been initially marked as "fissures", some may be marked as "pits", and some may be marked as multiple disease types at the same time, or not yet explicitly classified.

The computer device extracts the pending reference detection information and their corresponding road surface disease classifications from the database by executing specific query commands or data retrieval operations. This information will then be used in subsequent analysis and processing steps to help computer equipment more accurately identify and classify diseases on asphalt pavement.

For more specific illustration, assume that a computer device uses SQL (structured query language) to query its relational database. The computer device may execute an SQL command like the following: "SELECT FROM PendingInspections", wherein "PendingInspections" is a table name storing pending reference detection information. After executing this command, the computer device will return all records in the table, including the road surface fault classification information corresponding to each record.

The purpose of step S20b is to obtain the number Q4 corresponding to each road surface disease classification. This number Q4 represents the number of previously stored pending reference detection information belonging to a particular road surface disease class. Through the step, the computer equipment can know the distribution condition of various diseases in the data to be determined, and provides basis for subsequent analysis and decision. In executing step S20b, the computer device further processes the pending reference detection information acquired in step S20 a. Specifically, these pending information are traversed and the road surface fault classification associated with each piece of information is counted. This counting is accomplished by counting the amount of pending information for each disease classification. For example, it is assumed that in step S20a, the computer device acquires 1000 pieces of pending reference detection information covering various types of diseases that may occur on the asphalt pavement, such as cracks, pits, water damage, and the like. In step S20b, the computer device looks at the 1000 pieces of information one by one and counts the disease classifications that are marked by each piece of information. Finally, the computer device obtains the number Q4 corresponding to each disease classification, for example, 300 pieces of pending information for crack diseases, 200 pieces of pending information for pit diseases, and so on.

This number Q4 reflects the frequency of occurrence of various kinds of road surface diseases in the pending data. By comparing the number of different disease classifications, Q4, the computer device can determine which disease types are more common and which are likely to be rare. The method has important guiding significance for subsequent pavement disease identification and analysis work. For example, a disease type with a larger number Q4 may be the subject of significant attention by computer equipment, while a disease type with a smaller number Q4 may require more data collection or verification to ensure accuracy of its identification.

The step S20c aims to screen out the disease classifications meeting the specific number threshold value from the plurality of road surface disease classifications, and uses them as the key objects of the subsequent analysis, namely, to set the road surface disease classifications.

In performing this step, the computer device first sets a number threshold. This threshold is an important reference criterion for measuring whether the frequency of occurrence of a certain road surface disease classification in the pending reference test information has reached the point of interest. The setting of the threshold may be adjusted according to actual conditions, such as historical data, empirical determinations, or specific business requirements. The computer device then compares the number Q4 for each road surface fault classification to the number threshold. If the number Q4 of a disease classification is greater than (or equal to) the number threshold, as determined by the specific requirements, then this disease classification is considered an important classification that requires special attention, i.e., setting a road surface disease classification.

For example, assuming that the computer device sets the number threshold to 200, it means that a certain road surface disease classification is regarded as a set road surface disease classification only when it occurs more than 200 times in the pending reference detection information. According to the number Q4 acquired in step S20b, if the number Q4 of "crack" type diseases is 300, which is greater than the number threshold 200, then "crack" is selected as a set road surface disease classification. In contrast, if the number Q4 of "water damage" type diseases is only 150, and the number threshold is not reached, it is not selected as the setting of the road surface disease classification in step S20 c. In this way, the computer device can screen out road surface disease classifications which frequently occur in pending data and may be representative or important, and provide more targeted guidance for subsequent road surface disease identification, analysis and processing work. The method is not only beneficial to improving the accuracy of disease identification, but also improves the efficiency and performance of the whole computer equipment.

Step S300: and acquiring an integrated characterization vector of the asphalt pavement radar imaging image according to the asphalt pavement radar imaging image and the candidate pavement disease type information.

The purpose of step S300 is to obtain an integrated token vector for the image. The integrated characterization vector not only contains the semantic features of the image, but also fuses the disease type features, so that richer and more accurate information is provided for subsequent disease identification and analysis.

In performing step S300, the computer device first reviews and analyzes raw data of the asphalt pavement radar imaging image. The data are in the form of images, each pixel carrying reflective information of a portion of the road surface. The computer device extracts semantic features from the image by a specific algorithm, such as Convolutional Neural Network (CNN). These semantic features may include texture, structure, texture, etc. of the pavement, which are the basis for understanding the image content.

Meanwhile, the computer apparatus refers also to the candidate road surface disease type information determined in step S200. These information indicate the type of disease that may be present in the image, such as cracks, potholes, etc. Each disease type has its unique visual appearance and characteristics. Next, the computer device fuses the semantic features of the image with the disease category features. This process may be accomplished in a variety of ways, such as using feature stitching, weighted fusion, or automatic learning through a deep learning model. The result after fusion is an integrated characterization vector, which contains the whole information of the image and highlights the key features related to diseases.

By way of specific example, assume that a computer device is processing a radar imaging image of an asphalt pavement containing crack defects. The computer device first extracts semantic features of the image through the CNN, which may describe the roughness, continuity, etc. properties of the road surface. Then, the computer device combines the feature description about the crack, such as the shape, length, width, etc. of the crack in the candidate pavement defect type information, and fuses the feature with the semantic feature. The integrated characterization vector obtained finally can describe the radar image containing the crack disease comprehensively and accurately. The integrated characterization vector plays an important role in subsequent tasks such as disease identification, classification, positioning and the like. The method provides rich information for computer equipment to analyze and judge, thereby improving the accuracy and efficiency of disease detection.

Step S400: performing disease detection on the asphalt pavement radar imaging image according to the integrated characterization vector to obtain a predicted disease detection result of the asphalt pavement radar imaging image; the estimated disease detection result comprises estimated confidence degrees of the asphalt pavement radar imaging image corresponding to the plurality of set pavement disease classifications, and the estimated disease detection result is used for determining the set pavement disease classifications corresponding to the asphalt pavement radar imaging image.

And step S400, performing disease detection on the asphalt pavement radar imaging image by using the integrated characterization vector, and obtaining a predicted disease detection result. The step not only relates to the image processing technology, but also fuses the application of a machine learning model so as to realize the automatic identification and classification of diseases.

In performing step S400, the computer device first obtains pre-processed radar imaging images of the asphalt pavement, which have been subjected to a series of preprocessing steps (e.g., denoising, enhancement, etc.) to improve the image quality and highlight the disease features. The computer device then extracts the characteristic information in these images, which is an important basis for disease identification.

The computer device then analyzes the extracted feature information using a previously trained machine learning model (e.g., support vector machine, neural network, etc.). The model has learned the characteristic and classification rule of a large number of road surface diseases in the training stage, so that whether the diseases exist in the image and the type and degree of the diseases can be judged according to the input characteristic information.

In the process of model analysis, an integrated characterization vector is generated, and the vector fuses various characteristic information in the image, so that disease conditions in the image can be comprehensively reflected. Then, the computer equipment performs disease detection on the asphalt pavement radar imaging image according to the integrated characterization vector to obtain a predicted disease detection result.

The estimated disease detection result comprises estimated confidence that the asphalt pavement radar imaging image corresponds to the classification of various set pavement diseases. The estimated confidence is a numerical index used for measuring the judging and grasping degree of the computer equipment on whether a certain disease type exists in the image. For example, if the computer device determines that there is a high likelihood of crack disease in an image, the corresponding estimated confidence will be high. Finally, the estimated disease detection result is used for determining the set pavement disease classification corresponding to the asphalt pavement radar imaging image. The computer equipment judges the disease type in the image according to the estimated confidence and a preset classification threshold value, so that the whole disease detection process is completed.

For example, assume that a computer device acquires an asphalt pavement radar imaging image and extracts characteristic information therefrom. The computer device then analyzes the feature information using the trained neural network model to generate an integrated token vector. According to the vector, the computer equipment judges that the estimated confidence coefficient of the crack disease existing in the image is 0.9 (the value range is 0 to 1) which is far higher than the estimated confidence coefficient of other disease types. Thus, the computer device ultimately determines that the set pavement defect corresponding to this image is classified as "crack".

As an implementation manner, step S400, performing disease detection on the asphalt pavement radar imaging image according to the integrated characterization vector to obtain a predicted disease detection result of the asphalt pavement radar imaging image, may specifically include:

Step S410: extracting projection characterization vectors of a plurality of channels according to the integration characterization vectors;

Step S420: according to the integrated characterization vector, obtaining an influence coefficient set of each type of set pavement disease classification, wherein the influence coefficient set comprises influence coefficients of the channels, and the influence coefficients represent the contribution degree of characterization information of an image on the channel to whether the image is the set pavement disease classification;

Step S430: respectively carrying out weight average on projection characterization vectors of the channels according to a plurality of influence coefficients in the influence coefficient set of each type of set pavement disease classification to obtain a type characterization vector of each type of set pavement disease classification;

step S440: and respectively carrying out disease detection on the asphalt pavement radar imaging image according to the type characterization vector of each type of set pavement disease classification to obtain the estimated confidence corresponding to each type of set pavement disease classification.

Specifically, step S410 extracts a projected token vector for a plurality of channels from the integrated token vector, including but not limited to semantics, texture, color, and reference confidence. Projection characterization vectors, also known as mapping features, are specific numerical representations of the feature information on each channel, which provide an important basis for subsequent classification of road surface defects. In performing this step, the computer device first obtains the integrated token vector. This vector is a high-dimensional data structure in which various characteristic information extracted from the asphalt pavement radar imaging image is fused. Such characteristic information may include gray values, edge contours, frequency components, etc. of the image, which together constitute a comprehensive description of the image.

The computer device then decomposes the integrated token vector, mapping it onto different channels. Each channel corresponds to a specific feature type, such as semantic channels focusing on semantic information in the image, namely objects in the image and interrelationships thereof; the texture channel focuses on the texture characteristics of the image, namely the local mode and the arrangement rule of the image; the color channel extracts color information of the image, including hue, saturation, brightness and the like; the reference confidence channel reflects the credibility of each region in the image, and is generally used for measuring the accuracy of image segmentation or identification. During the mapping process, the computer device may use a series of mathematical transformations and algorithms to process the integrated token vector. For example, on a semantic channel, a computer device may employ word embedding techniques to convert tags or descriptions in an image into vector form; on the texture channel, the computer device may use a filter or convolutional neural network to extract texture features of the image; on the color channels, the computer device may convert the color space (e.g., RGB) of the image to a color space (e.g., HSV) more suitable for analysis.

Ultimately, each channel generates a projected token vector that together form a comprehensive description of the image in the various feature dimensions. These projected characterization vectors will be used for subsequent road surface fault classification and identification tasks to improve the accuracy and reliability of the detection.

For example, assume that a computer device is processing a radar imaging image of an asphalt pavement containing crack defects. In step S410, the computer device first extracts the integrated token vector of the image. Then, on the semantic channel, the computer device can identify the keyword "crack" in the image and convert it into a corresponding vector representation; on the texture channel, the computer device may extract edge contours and texture features of the fracture region; on the color channel, the computer device may notice that the color of the crack area is different from the surrounding pavement; whereas on the reference confidence channel, the computer device may evaluate the confidence level of the fracture region based on its sharpness and contrast. These projected characterization vectors will provide strong support for subsequent classification of road surface defects.

Next, step S420 determines the influence coefficients of different types of road surface diseases on the respective characteristic channels, thereby guiding the identification and classification of the diseases more precisely. In this step, the computer device needs to analyze the integrated characterization vector and calculate the set of influence coefficients of the integrated characterization vector on different channels (such as semantics, texture, color, reference confidence, etc.) for each type of preset road surface disease.

First, the integrated token vector is a comprehensive description that integrates multiple features of an image. It may be a high-dimensional array containing information of the features of color, texture, shape, etc. extracted from the image. This information may have different importance for different types of road surface diseases and therefore needs to be considered separately.

Next, a set of influence coefficients will be determined for each set of road surface defects (e.g., cracks, pits, oil flooding, etc.). This set includes the influence coefficients, i.e. weights, of the multiple channels. Each weight represents the contribution degree of a characteristic channel when judging whether the image belongs to the disease. For example, for a crack failure, the texture channel may contribute significantly, as the presence of a crack may significantly alter the texture characteristics of the image; for oil-spreading diseases, the color channel may be more important, as oil-spreading typically results in a change in road color.

In determining these impact coefficients, the computer device may employ a machine learning approach. Specifically, the importance of different feature channels for disease classification can be learned by training a classifier (such as a support vector machine, a neural network, etc.) by using image samples and corresponding labels of known road surface disease classifications. In the training process, the weights of all channels are continuously adjusted according to the performance of the classifier until the optimal classifying effect is achieved. And finally, the influence coefficient set of each type of pavement diseases obtained through the step provides important basis for subsequent disease detection and classification. The computer equipment can utilize the coefficients to adjust the weights of different characteristic channels when judging whether the image belongs to a certain disease, so that the accuracy and the efficiency of disease identification are improved.

For example, in one practical application scenario, assume that the computer device has obtained projection characterization vectors for multiple channels of an image via step S410. Now, in step S420, the computer device needs to determine whether this image belongs to the class of road surface defects of "cracks". According to the influence coefficient set of the crack disease obtained through training in advance, the computer equipment knows that the importance of the texture channel for judging the crack is higher. Therefore, in subsequent calculations, the computer device gives the projected token vector of the texture channel a greater weight, thereby more accurately determining whether the image contains a crack defect. Then, step S430 performs weighting processing on the projection characterization vectors of the plurality of channels to obtain a type characterization vector for each type of set road surface disease. The core of this step is to use the set of influence coefficients calculated in the previous step, which reflect the contribution, i.e. the importance, of the different channels in judging the specific disease type. In performing this step, the computer device first reviews the set of influence coefficients for each set of road surface fault classifications obtained in step S420. This set contains the influence coefficients, i.e. weights, of a number of channels (e.g. semantics, texture, color, reference confidence, etc.). These weights are derived from a large number of training data and machine learning algorithms and represent the contribution of each channel in identifying a particular disease.

Next, the projection characterization vectors of the plurality of channels obtained in step S410 are weighted by using the weights in the corresponding influence coefficient sets for each type of road surface diseases set. Specifically, the projection characterization vector of each channel is multiplied by a corresponding weight, and this process can be understood as emphasizing or suppressing the feature information according to the importance of the channel.

For example, assume that a "crack" is being processed for this type of lesion, and that projected token vectors for four channels of semantics, texture, color, and reference confidence have been obtained through the previous steps. In step S430, the computer device uses the set of influence coefficients for the "crack" defect, wherein the weight of the possible texture channels is higher, because texture features are particularly important for identifying cracks. The computer device then multiplies the projected token vector for the texture channel by a larger weight value and the projected token vectors for the other channels by a relatively smaller weight value.

After the weighting process is completed, the computer device performs an averaging operation on all weighted projection token vectors to obtain a type token vector integrating the information of each channel. The type characterization vector is a comprehensive description of specific disease types, integrates information of images in a plurality of characteristic dimensions, and provides powerful support for subsequent disease detection and classification.

Through the processing of step S430, the computer device can more accurately capture the key characteristics of each type of set pavement diseases, thereby improving the accuracy and reliability of disease detection. The weight average-based method effectively utilizes complementary information among a plurality of channels, so that computer equipment can make more accurate judgment when facing complex and changeable asphalt pavement diseases.

Step S440 is to use the type characterization vector calculated in the previous step to detect the disease of the actual asphalt pavement radar imaging image, and output the estimated confidence of each type of set pavement disease. The main purpose of this step is to determine if a specific disease is present in the image and to give a degree of confidence in the determination.

In performing this step, the computer device first reviews the type characterization vector for each type of set pavement defect classification obtained in step S430. These types of token vectors are feature descriptions that integrate multiple channels (e.g., semantic, texture, color, reference confidence, etc.) of information, which represent unique manifestations of various diseases in a feature space. Next, using these types of characterization vectors as references, disease detection is performed on the input asphalt pavement radar imaging image one by one. The process of detection can be seen as finding a feature pattern in the image that matches the type characterization vector. To achieve this, the computer device may employ a machine learning model, such as a Support Vector Machine (SVM), random Forest (Random Forest), or deep learning network (such as convolutional neural network CNN), which has been trained to identify features associated with a particular disease type.

For example, assuming that the type of disease "crack" is being detected, the various regions of the asphalt pavement radar imaging image are compared to a type characterization vector for the "crack" disease. If the characteristics of a certain area are highly similar to the type characterization vector of a crack, the computer equipment judges that crack diseases exist in the area, and calculates an estimated confidence coefficient to represent the reliability of the judgment. The computation of the confidence in the predictions is typically based on the degree of feature matching and the output of the machine learning model. It may be a number between 0 and 1, where 0 means completely uncertain and 1 means completely deterministic. In practical applications, the computer device may assign an estimated confidence level to each detected disease instance, so that the user or the subsequent processing flow may know the reliability of the detection result.

Through the processing of step S440, the computer device can output a detection result including various set road surface diseases and their corresponding estimated confidence levels. The method not only helps users to know the disease condition of the asphalt pavement, but also provides decision support for subsequent maintenance and repair work.

As an implementation manner, in step S400, after performing disease detection on the asphalt pavement radar imaging image according to the integrated characterization vector to obtain the estimated disease detection result of the asphalt pavement radar imaging image, the method may further include:

step S500: taking the set pavement disease classification with the maximum estimated confidence as the set pavement disease classification corresponding to the asphalt pavement radar imaging image;

Step S600: determining a correlation coefficient between the asphalt pavement radar imaging image and each piece of undetermined reference detection information according to the set pavement disease classification corresponding to the asphalt pavement radar imaging image and the set pavement disease classification corresponding to the plurality of undetermined reference detection information corresponding to the asphalt pavement radar imaging image;

Step S700: and taking the undetermined reference detection information with the association coefficient meeting the association coefficient requirement as target detection information determined according to the asphalt pavement radar imaging image.

In step S500, the computer device compares the estimated confidence of each set road surface disease classification obtained in step S440. The estimated confidence is a quantitative representation of the degree of confidence that a computer device is in the presence of a particular disease type for a region. In this step, the computer device selects the set road surface fault classification with the highest estimated confidence as the set road surface fault classification corresponding to the asphalt pavement radar imaging image. For example, if the estimated confidence of the computer device for a "crack" defect is 0.9 and the estimated confidence of the computer device for a "pit" defect is 0.7, the computer device may determine that the type of defect corresponding to the image is "crack".

Step S600 involves determining a correlation coefficient between the asphalt pavement radar imaging image and each of the pending reference detection information. The correlation coefficient is an index that measures the correlation between the image and the reference detection information. To determine this coefficient, the computer device may use a relevance model. This model may be a pre-trained machine learning model, such as a Support Vector Machine (SVM), a logistic regression model, or a deep learning based model, such as a Convolutional Neural Network (CNN). The model input includes image features and corresponding reference detection information features, and the output is a correlation coefficient. For example, the computer device may extract features such as texture, shape, etc. of the image, compare the extracted features with corresponding features in the reference detected information, and calculate a correlation coefficient between 0 and 1 through the model, where 1 indicates complete correlation and 0 indicates no correlation.

In practice, the computer device may input the feature vector of the asphalt pavement radar imaging image and the feature vector of each of the pending reference detection information into the correlation model. The model outputs a correlation coefficient by comparing the similarity of the feature vectors. For example, if a certain pending reference detection information describes a crack that is highly similar to the type, location, and size of disease in the image, the correlation model will output a correlation coefficient that is close to 1.

In step S700, the computer device determines which pending reference detection information can be used as a final basis according to the association coefficient calculated in step S600. The computer device sets a threshold value for the correlation coefficient and only when the correlation coefficient between the pending reference detection information and the image exceeds the threshold value, the reference detection information, i.e. the target detection information, is considered to be valid. For example, the computer device may set the association coefficient threshold to 0.8, and then all pending reference detection information having an association coefficient greater than or equal to 0.8 may be selected as the final reference detection information. These ultimately selected reference detection information will provide more comprehensive and accurate disease analysis and decision support for the user along with disease detection results of the asphalt pavement radar imaging image.

As an implementation manner, in the computer device provided by the embodiment of the present application, a radar imaging image detection network is deployed in advance, and is used for performing disease detection of an asphalt pavement radar imaging image, and specifically, the radar imaging image detection network includes a characterization vector extraction component and a disease detection component. Step S300, wherein the obtaining the integrated characterization vector of the asphalt pavement radar imaging image according to the asphalt pavement radar imaging image and the candidate pavement disease type information may include: and acquiring the integrated characterization vector according to the asphalt pavement radar imaging image and the candidate pavement disease type information through the characterization vector extraction component. Step S400, performing disease detection on the asphalt pavement radar imaging image according to the integrated characterization vector to obtain a predicted disease detection result of the asphalt pavement radar imaging image, which may include: and performing disease detection on the asphalt pavement radar imaging image according to the integrated characterization vector by the disease detection assembly to obtain the estimated disease detection result.

In the above embodiment, the computer device is provided with a special radar imaging image detection network which is designed to perform the disease detection task of the asphalt pavement radar imaging image. The network structure comprises two main components: a characterization vector extraction component and a disease detection component, which are respectively responsible for processing the different phases of the image. The characterization vector extraction component is the first link in the radar imaging image detection network, and the main function of the characterization vector extraction component is to analyze and extract the characteristic information of the input image. In performing step S300, the component receives as inputs the asphalt pavement radar imaging image and the candidate pavement defect type information. It then uses deep learning techniques such as Convolutional Neural Networks (CNNs) to extract key features in the image, such as texture, shape, edges, etc., and encodes these features into a high-dimensional vector form, i.e., an integrated token vector. The vector is a mathematical description of the image content, captures critical information related to diseases in the image, and provides a basis for subsequent detection. For example, if a crack defect is included in the input asphalt radar imaging image, the token vector extraction component can identify the presence of the crack and encode its features (e.g., crack length, width, shape, etc.) into the integrated token vector.

The disease detection component is a second link in the radar imaging image detection network, and utilizes the integrated characterization vector output by the characterization vector extraction component to detect the disease. In performing step S400, the disease detection component receives as input the integrated characterization vectors and classifies and identifies these vectors via a machine learning algorithm (e.g., support Vector Machine (SVM), decision tree, neural network, etc.). In this process, the disease detection component references pre-trained models that have learned how to determine whether a particular disease type is present in the image based on the feature vectors. Continuing with the example above, the lesion detection assembly receives an integrated token vector comprising a fracture feature and determines, via a classification algorithm, whether the image region corresponding to the vector is indeed afflicted with a fracture. If so, it outputs a predicted disease detection result, which may include information about the type, location, and severity of the disease.

In this way, the computer equipment can automatically complete the disease detection task of the asphalt pavement radar imaging image, and the detection efficiency and accuracy are greatly improved.

Based on this, as an embodiment, the radar imaging image detection network may be obtained by performing debugging by:

Step S10: and acquiring a first asphalt pavement radar imaging image sample and a plurality of sample reference detection information corresponding to the first asphalt pavement radar imaging image sample, wherein the sample reference detection information comprises confirmed past reference detection information corresponding to the first asphalt pavement radar imaging image sample.

In this step, the computer device needs to acquire two types of critical data, namely a first asphalt pavement radar imaging image sample and sample reference detection information corresponding to the image samples.

The first asphalt pavement radar imaging image sample is an actual scanned asphalt pavement image, which represents different pavement conditions and possible disease conditions. These image samples will be used as training data for constructing and optimizing the radar imaging image detection network. The sample reference detection information is a past disease detection result which is associated with the image samples and has been confirmed. These information are obtained by historical detection activities, which reflect the disease condition of the target road segment at different points in time. In particular, the sample reference detection information may include detailed data of disease location, type, severity, etc., which provide accurate labels and references for the training process. In the training process, the computer equipment takes the image samples and corresponding sample reference detection information as input, so that the network learns how to identify and extract pavement disease features from radar images. By comparing the differences between the predicted results of the network and the sample reference detection information, the computer device can evaluate the performance of the network and adjust and optimize it as needed.

The content of step S10 may refer to the aforementioned step S100.

Step S20: and determining first candidate pavement disease type information of the first asphalt pavement radar imaging image sample according to set pavement disease classifications corresponding to the plurality of sample reference detection information, wherein the first candidate pavement disease type information comprises sample candidate confidence coefficients of the first asphalt pavement radar imaging image sample corresponding to the plurality of set pavement disease classifications.

The content of step S20 may refer to step S200 described above, and will not be described here.

Step S30: and acquiring a sample integration characterization vector of the first asphalt pavement radar imaging image sample according to the first asphalt pavement radar imaging image sample and the first candidate pavement disease type information through the characterization vector extraction component.

In step S30, the computer device uses a token vector extraction component that is capable of outputting an integrated token vector based on the input image sample and candidate road surface disease category information. For example, the computer device may input the first asphalt pavement radar imaging image sample and the first candidate pavement defect type information together into the characterization vector extraction component. This component may be a machine learning model, such as a Deep Convolutional Neural Network (DCNN), which has been trained to extract features from images that are related to road surface damage. Inside the token vector extraction component, the input image is subjected to a series of preprocessing operations, such as scaling, normalization, etc., to ensure consistency and validity of the image data. Then, deep feature extraction is performed on the image by using structures such as a convolution layer and a pooling layer. These features may include information on the texture, shape, edges, etc. of the image, which are critical for identifying road surface imperfections.

At the same time, the characterization vector extraction component also obtains a characteristic representation of the first candidate pavement disease category information, such as a unique visual representation and characteristics of each disease type, which may be predetermined a priori characteristics.

The characterization vector extraction component outputs an integrated characterization vector, which is a result of fusing the pavement disease features in the first asphalt pavement radar imaging image sample with the features of the first candidate pavement disease type information, for example, completing the stitching. This characterization vector will serve as an important input for subsequent model training and evaluation, helping computer equipment to more accurately identify and evaluate various diseases on asphalt pavement.

The content of step S30 may refer to step S300.

As an embodiment, the feature vector extraction component includes a first extraction unit, a second extraction unit, and a feature information interaction unit, based on which, in step S30, the obtaining, by the feature vector extraction component, a sample integration feature vector of the first asphalt pavement radar imaging image sample according to the first asphalt pavement radar imaging image sample and the first candidate pavement disease type information may specifically include: extracting, by the first extracting unit, a sample image characterization vector of the first asphalt pavement radar imaging image sample; extracting sample disease information characterization vectors of the first candidate pavement disease type information through the second extraction unit; and carrying out interactive integration on the sample image characterization vector and the sample disease information characterization vector through the characterization information interaction unit to obtain the sample integration characterization vector.

The characterization vector extraction component consists of a first extraction unit, a second extraction unit and a characterization information interaction unit, wherein each unit has a specific function.

First, a sample image characterization vector of a first asphalt pavement radar imaging image sample is extracted by a first extraction unit. In this step, the first decimation unit may be performed using a Convolutional Neural Network (CNN) in deep learning. The convolutional neural network can automatically learn the hierarchical feature expression in the image, and is very effective for the image recognition task. For example, assuming that the first extraction unit employs a pre-trained CNN model, such as VGG16 or ResNet, which has been trained on a large amount of image data, key features in the image can be extracted efficiently. When the first asphalt pavement radar imaging image sample is input to the first extraction unit, the CNN model carries out convolution operation, pooling operation and the like on the first asphalt pavement radar imaging image sample, and finally outputs a sample image representation vector representing image characteristics. This vector captures key information in the image, such as texture, shape, structure, etc., and provides a basis for subsequent pavement defect identification.

Next, the sample disease information characterization vector of the first candidate road surface disease type information is extracted by the second extraction unit. In this step, the second extraction unit may use Natural Language Processing (NLP) technology or a specific embedding layer to convert the disease type information into a vector form. For example, if the first candidate road surface fault category information is given in text form, such as crack size, pit size, and distribution location, etc., the second extraction unit may use Word embedding techniques, such as Word2Vec or GloVe, to convert these text labels into a high-dimensional vector representation. These vectors capture the semantic relationships between disease species, helping the model understand the similarity and variability between different diseases. Of course, in one embodiment, since the road surface disease type information is determined, in order to save time and cost and calculation power, the characteristic information of the determined road surface disease type information may be extracted in advance and stored, and directly retrieved when needed, where the second extraction component may be a retrieval module.

And finally, carrying out interactive integration on the sample image characterization vector and the sample disease information characterization vector through a characterization information interaction unit to obtain a sample integration characterization vector. In this step, the token information interaction unit may use techniques such as attention mechanism and feature fusion to implement vector interaction and integration. For example, the two token vectors are directly spliced to obtain an integrated token vector. Or adopting an attention mechanism, wherein the attention mechanism can dynamically adjust the importance of different features in the sample image characterization vector according to the content of the sample disease information characterization vector, so that the model can pay more attention to the image features related to the current disease type. Through the interactive integration mode, the sample integration characterization vector not only contains the characteristic information of the image, but also integrates the priori knowledge of the disease types, so that the performance of the model on the road surface disease identification task is improved.

And sequentially executing extraction and integration operations through three units of the characterization vector extraction component to finally obtain a sample integration characterization vector fused with image characteristics and disease type information. This vector will serve as an important input for subsequent model training and evaluation, helping computer equipment to more accurately identify and evaluate various diseases on asphalt pavement.

Step S40: and performing disease detection on the first asphalt pavement radar imaging image sample according to the sample integration characterization vector through the disease detection assembly to obtain a first estimated disease detection result, wherein the first estimated disease detection result comprises sample estimated confidence degrees of the first asphalt pavement radar imaging image sample corresponding to the plurality of set pavement disease classifications.

The computer device inputs the sample integration characterization vector obtained in step S30 into the disease detection component. This component may be a specially trained machine learning model, such as Support Vector Machine (SVM), softmax, fully connected network, etc., for classification and recognition tasks. Models have been trained on a large number of labeled data to learn how to identify various road surface defects from the input characterization vectors. Inside the disease detection component, the model performs a series of processes and analyses on the input sample integration characterization vector. By comparing the similarity of the input vector and various disease feature vectors in the training data, the model can determine the type of disease that may be present in the input image. Meanwhile, according to the characteristic distribution and the mode of the input vector, the existence possibility of each disease is evaluated, and an estimated confidence is given.

For example, assume that the disease detection component employs a deep neural network model that learns during training how to identify road surface disease based on an input characterization vector. When a sample integration characterization vector of a first asphalt pavement radar imaging image sample is input, the model firstly extracts key features in the vector, such as features related to cracks, pits and other diseases. The model then compares these features to features in the training data to find the most similar disease classification. Finally, the model gives an estimated confidence level of the possibility of existence of the disease classification according to the matching degree and the distribution mode of the features.

Through the processing of step S40, the computer device can output a first estimated disease detection result. The result not only comprises the possible disease types in the first asphalt pavement radar imaging image sample, but also gives the estimated confidence of each disease. For example, the results may show that the estimated confidence of the presence of a crack in the image is 0.85, the estimated confidence of the presence of a pit is 0.7, and so on. The information provides an important reference basis for subsequent pavement disease evaluation and repair.

As one embodiment, the disease detection component comprises a characterization vector projection unit corresponding to a plurality of channels, a weight adjustment unit for setting road surface disease classification of each class and a disease identification unit. Based on this, step S40, through the disease detection component, performs disease detection on the first asphalt pavement radar imaging image sample according to the sample integration characterization vector, to obtain a first estimated disease detection result, which may specifically include:

step S41: respectively integrating the characterization vectors according to the samples by a plurality of characterization vector projection units, and extracting sample projection characterization vectors of the channels;

Step S42: the method comprises the steps that a weight adjusting unit of each type of set pavement disease classification is used for integrating a characterization vector according to the sample, and an influence coefficient set of each type of set pavement disease classification is obtained, wherein the influence coefficient set comprises influence coefficients of a plurality of channels, and the influence coefficients represent the contribution degree of characterization information of an image on the channels to whether the image is the set pavement disease classification or not;

Step S43: respectively carrying out weight average on sample projection characterization vectors of the channels according to a plurality of influence coefficients in an influence coefficient set of each type of set pavement disease classification by a weight adjusting unit of each type of set pavement disease classification to obtain sample type characterization vectors of each type of set pavement disease classification;

Step S44: and respectively carrying out disease detection on the first asphalt pavement radar imaging image samples according to the sample type characterization vectors of each type of set pavement disease classification by the disease identification unit of each type of set pavement disease classification to obtain the sample prediction confidence corresponding to each type of set pavement disease classification.

The disease detection assembly is composed of a plurality of parts, including a characterization vector projection unit, a weight adjustment unit and a disease identification unit, each part having a specific function thereof.

First, step S41 is to extract sample projection token vectors of a plurality of channels by a plurality of token vector projection units. These projection units may be expert networks in MMOE's structure, each of which focuses on extracting channel-specific information from the sample-integrated token vector. For example, if there are three channels, then there are three expert networks that extract features associated with the three channels, respectively. These features may include texture, shape, edges, etc., which are critical to subsequent identification of road surface imperfections. The result of each expert network output is a sample projection characterization vector that captures the key information of the input vector on a particular channel.

Next, step S42 is to acquire an influence coefficient set by the weight adjustment unit that sets the road surface disease classification for each class. These weight adjustment units can be analogous to the Gate network in MMOE, whose role is to dynamically adjust the importance of the different channels according to the input vector. Specifically, for each set road surface fault classification (e.g., cracks, pits, etc.), there will be a corresponding gate network. The gate network calculates an influence coefficient set according to the sample integration characterization vector, and each influence coefficient in the set represents the contribution degree of characterization information on a channel to judging whether the image belongs to the disease. For example, if a feature on a channel is particularly important for identifying a crack, the coefficient of influence corresponding to that channel may be relatively high.

Then, step S43 is to use the set of influence coefficients of each set of road surface disease classifications to weight average the sample projection characterization vectors of the plurality of channels, thereby obtaining the sample type characterization vector of each set of road surface disease classifications. In practice, the information of each channel is integrated, and the characteristics of different channels are fused together in a weighted average mode to form a more comprehensive and targeted characterization vector. This vector contains not only the key features of the image on each channel, but also the importance of these features for identifying a particular disease.

Finally, step S44 performs disease detection on the first asphalt pavement radar imaging image samples by using each type of disease recognition unit for setting pavement disease classification, and outputs the sample prediction confidence. These lesion recognition units may be deep learning based classifiers, such as fully connected layer plus softmax activation function structures. The disease judgment is carried out according to the sample type characterization vector obtained in the previous step, and an estimated confidence is given to indicate the possibility that the image belongs to a certain disease. For example, if a certain disease recognition unit judges that there is a crack in an image with a confidence of 0.9, it means that the image is 90% likely to contain a crack.

The operations of projection, weight adjustment, weighted average, disease identification and the like are sequentially carried out by a plurality of units of the disease detection assembly, and finally a first estimated disease detection result containing various set road surface disease classification estimated confidence degrees is obtained. The result not only provides an important reference for subsequent pavement disease evaluation and repair, but also verifies the effectiveness and accuracy of the computer equipment in the network debugging process.

Step S50: according to a first priori disease detection result and the first estimated disease detection result of the first asphalt pavement radar imaging image sample, the radar imaging image detection network is debugged, so that errors between the first estimated disease detection result and the first priori disease detection result obtained by the radar imaging image detection network after debugging are reduced, and the first priori disease detection result comprises actual confidence that the first asphalt pavement radar imaging image sample corresponds to the classification of the various set pavement diseases.

Step S50 optimizes and debugs the radar imaging image detection network based on the two disease detection results to reduce the error between the predicted result and the actual result. Specifically, this step involves two key disease detection results: the first priori disease detection result and the first estimated disease detection result.

The first priori disease detection results are obtained based on the existing and verified detection method or computer equipment, and represent the actual existence of various set pavement disease classifications in the first asphalt pavement radar imaging image sample. This result includes actual confidence levels corresponding to the various set road surface fault classifications that reflect the likelihood that each fault will actually exist in the image. For example, for an image containing cracks and pits, the a priori detection result may give an actual confidence of 0.9 for the presence of cracks and 0.8 for the presence of pits.

The first estimated disease detection result is obtained through preliminary analysis of a radar imaging image detection network, and represents the prediction of the type and degree of the possible disease in the image by the network. This result also includes estimated confidence levels for the various set road surface fault classifications reflecting the likelihood that each fault is considered by the network to be present in the image.

In step S50, the computer device compares the two disease detection results, in particular the confidence differences between them. The computer device then adjusts the parameters and structure of the radar imaging image detection network based on these differences to reduce errors in the network during the prediction process. Such tuning may include adjusting weights, biases, number of layers, number of neurons, etc. of the network, or optimizing learning algorithms and training strategies of the network. For example, if the estimated confidence of a network for a disease class is always lower than the actual confidence, then the computer device may increase the weight of features associated with the disease class in the network, or increase the network layer dedicated to identifying the disease class, to increase the sensitivity of the network to the disease class.

Finally, through repeated debugging and optimization, the computer equipment can enable the error between the result of the radar imaging image detection network and the first priori disease detection result to be gradually reduced when the first estimated disease detection result is output, so that the accuracy and the reliability of the network are improved. This process is of great importance to ensure the effective application of radar imaging technology in road surface disease detection.

As an implementation manner, the step S50 is configured to debug the radar imaging image detection network according to the first priori disease detection result and the first estimated disease detection result of the first asphalt pavement radar imaging image sample, so that an error between the first estimated disease detection result obtained by the radar imaging image detection network after the debugging and the first priori disease detection result is reduced, and includes:

step S51: according to cost influence coefficients corresponding to each set pavement disease classification, carrying out weight average on cross entropy between the estimated confidence coefficient and the actual confidence coefficient of the sample of each set pavement disease classification to obtain a cost parameter, wherein the degree of commonality between the estimated confidence coefficient and the actual confidence coefficient of the sample is inversely related to the cross entropy;

Step S52: and according to the cost parameter, debugging the radar imaging image detection network, and adjusting the cost influence coefficient, so that the cost parameter obtained according to the radar imaging image detection network after the debugging and the adjusted cost influence coefficient is reduced.

In step S51, a cost parameter is calculated. First, the computer device needs to determine cost impact coefficients corresponding to each set road surface disease classification, and the cost impact coefficients reflect the importance or the impact degree of different disease types in detection. For example, certain diseases may have a greater impact on road safety and should therefore be given a higher weight in the detection. The computer device then uses these cost impact coefficients to weight-fuse the cross entropy between the estimated confidence and the actual confidence for each class of disease samples. Cross entropy is an indicator of the difference between two probability distributions, and is used here to quantify the inconsistency between the predicted outcome and the actual outcome. Specifically, if the degree of commonality between the sample pre-estimated confidence and the actual confidence is higher, i.e., the closer the two are, the lower the cross entropy is; otherwise, if the difference between the two is larger, the cross entropy is higher, namely the degree of commonality between the estimated confidence and the actual confidence of the sample is inversely related to the cross entropy. Through weighting and fusing the cross entropy of different disease types, the computer equipment obtains a total cost parameter which comprehensively reflects the performance of the current radar imaging image detection network in various disease detection.

Step S52 is to debug the radar imaging image detection network based on the cost parameter calculated in step S51. The goal of the debugging is to reduce the value of the cost parameter by adjusting the parameters and structure of the network and adjusting the cost impact coefficient. This means that the computer device needs to find a network configuration and cost weight setting, so that under this configuration, the error between the estimated disease detection result and the priori disease detection result output by the network is minimized. This process may involve adjustments to the network layer number, neuron number, learning rate, etc. super parameters, as well as fine tuning of cost impact coefficients. Through repeated debugging and verification, the computer equipment can finally find an optimized network model and cost weight setting, so that the accuracy and reliability of the radar imaging image detection network in road surface disease detection are improved.

For example, assume that a computer device is processing an asphalt pavement radar imaging image sample containing both cracks and pits. In step S51, the computer device may assign a higher cost impact coefficient to the crack, because the crack may have a greater impact on road safety. Then, the computer equipment calculates the cross entropy between the estimated confidence coefficient and the actual confidence coefficient of two diseases of the crack and the pit, and uses the cost influence coefficient to carry out weighted fusion, so as to obtain an overall cost parameter. In step S52, the computer device may commission the radar imaging image detection network based on this cost parameter, such as increasing a network layer dedicated to identifying the fracture or increasing the weight of the fracture-related feature in the network. At the same time, the computer device will also fine tune the cost impact coefficients to find the optimal weight settings. Through the adjustment, the computer equipment can improve the performance of the radar imaging image detection network in the aspect of identifying pavement diseases such as cracks, pits and the like.

As an implementation manner, in the step S50, after the radar imaging image detection network is debugged according to the first priori disease detection result and the first estimated disease detection result of the first asphalt pavement radar imaging image sample, the method may further include:

Step S60: obtaining a second asphalt pavement radar imaging image sample, and taking set pavement disease classification information as second candidate pavement disease type information of the second asphalt pavement radar imaging image sample, wherein the second candidate pavement disease type information comprises sample candidate confidence coefficients of the second asphalt pavement radar imaging image sample corresponding to the plurality of set pavement disease classifications.

The second candidate road surface defect type information includes sample candidate confidence that the second asphalt road surface radar imaging image sample corresponds to a plurality of types of set road surface defect classification information that is the same as the set road surface defect classification information in the above embodiments. Because some second asphalt pavement radar imaging image samples may not have corresponding sample reference detection information, the second candidate pavement disease type information of the second asphalt pavement radar imaging image samples cannot be determined according to the set pavement disease classification corresponding to the sample reference detection information, the set pavement disease classification information is directly determined as the second candidate pavement disease type information of the second asphalt pavement radar imaging image samples, so that the asphalt pavement radar imaging image samples can train the radar imaging image detection network by using the asphalt pavement radar imaging image samples when the corresponding sample reference detection information is not available, and the scale of the asphalt pavement radar imaging image samples is increased.

Step S70: obtaining a sample integration characterization vector of the second asphalt pavement radar imaging image sample according to the second asphalt pavement radar imaging image sample and the second candidate pavement disease type information through the characterization vector extraction component, and performing disease detection on the second asphalt pavement radar imaging image sample according to the sample integration characterization vector through the disease detection component to obtain a second estimated disease detection result, wherein the second estimated disease detection result comprises the estimated confidence coefficient of the second asphalt pavement radar imaging image sample corresponding to the samples of the multiple set pavement disease classifications.

The second estimated disease detection result comprises the estimated confidence of the second asphalt pavement radar imaging image sample corresponding to the sample of the multi-class set pavement disease classification. For specific procedures reference is made to the previous steps.

Step S80: according to a second priori disease detection result and the second estimated disease detection result of the second asphalt pavement radar imaging image sample, the radar imaging image detection network is debugged, so that errors between the second estimated disease detection result obtained by the radar imaging image detection network after debugging and the second priori disease detection result are reduced, and the second priori disease detection result comprises actual confidence that the second asphalt pavement radar imaging image sample corresponds to the classification of the various set pavement diseases.

And obtaining a second priori disease detection result of the second asphalt pavement radar imaging image sample, wherein the second priori disease detection result comprises actual confidence levels of the second asphalt pavement radar imaging image sample corresponding to the multiple classes of set pavement disease classifications. For specific procedures reference is made to the previous steps. Optionally, the second priori disease detection result is manually annotated by an expert, for example, the actual confidence corresponding to the set pavement disease classification corresponding to the second asphalt pavement radar imaging image sample is marked as 1, the actual confidence corresponding to the rest set pavement disease classifications is marked as 0, and the actual confidence corresponding to the multiple set pavement disease classifications form the second priori disease detection result.

According to the embodiment of the application, the radar imaging image detection network is debugged based on two links, the steps S10-S50 are the first debugging link, the steps S60-S80 are the second debugging link, and the debugging sample cases of the two debugging links are different. In the first debugging link, according to the set pavement disease classification corresponding to the reference detection information of the multiple samples, first candidate pavement disease type information of a first asphalt pavement radar imaging image sample is determined. In the second debugging link, the set road surface disease classification information is used as second candidate road surface disease type information of a second asphalt pavement radar imaging image sample. For example, the confidence of the sample candidates for the plurality of set road surface deterioration classifications in the set road surface deterioration classification information is 50% or 100%.

The two debugging links are different in mode of determining the prior disease detection result, in the first debugging link, the first prior disease detection result of the first asphalt pavement radar imaging image sample is determined according to the first candidate pavement disease type information of the first asphalt pavement radar imaging image sample, for example, the actual confidence corresponding to the set pavement disease classification with the largest sample candidate confidence is set to be 1, and the actual confidence corresponding to the set pavement disease classification with the non-largest sample candidate confidence is set to be 0. In the second debugging link, the disease detection result marked for the second asphalt pavement radar imaging image sample is used as the second priori disease detection result of the second asphalt pavement radar imaging image sample, for example, the actual confidence corresponding to the set pavement disease classification corresponding to the second asphalt pavement radar imaging image sample is marked as 1, and the actual confidence corresponding to the other set pavement disease classifications is marked as 0. In the first debugging link, because the first asphalt pavement radar imaging image sample has high relation with the confirmed past reference detection information, the sample candidate confidence coefficient of the first asphalt pavement radar imaging image sample corresponding to multiple types of set pavement disease classifications can be approximately determined based on the set pavement disease classifications corresponding to the confirmed past reference detection information, the radar imaging image detection network is debugged based on the first asphalt pavement radar imaging image sample and the corresponding sample candidate confidence coefficient, the radar imaging image detection network can acquire the estimated confidence coefficient corresponding to the asphalt pavement radar imaging image according to the knowledge of the asphalt pavement radar imaging image and the corresponding candidate confidence coefficient, and the monitoring effect of the radar imaging image detection network is improved. In addition, in the first debugging link, the first priori disease detection result of the first asphalt pavement radar imaging image sample is determined according to the first candidate pavement disease type information, and the method has the advantages of being automatic and high in efficiency.

And in the second debugging link, because some second asphalt pavement radar imaging image samples possibly have no corresponding sample reference detection information, the second candidate pavement disease type information of the second asphalt pavement radar imaging image samples cannot be determined according to the set pavement disease classification corresponding to the sample reference detection information, and then the set pavement disease classification information is determined as the second candidate pavement disease type information of the second asphalt pavement radar imaging image samples, so that when the asphalt pavement radar imaging image samples have no corresponding sample reference detection information, the asphalt pavement radar imaging image samples can be used for debugging the radar imaging image detection network, and the scale of the asphalt pavement radar imaging image samples is increased. In addition, in the second debugging link, the second priori disease detection result of the second asphalt pavement radar imaging image sample is a manual annotation, the accuracy of the second priori disease detection result is high, the radar imaging image detection network is debugged again based on the second priori disease detection result, the radar imaging image detection network is finished to be refined, and the accuracy of the radar imaging image detection network is improved again.

In the embodiment of the application, a radar imaging image detection network for detecting the disease of the radar imaging image of the asphalt pavement is debugged, the radar imaging image detection network comprises a disease detection component corresponding to each type of set pavement disease classification, each disease detection component carries out disease detection according to an integrated characterization vector obtained by weighting projection characterization vectors output by a plurality of characterization vector projection units so as to obtain the confidence coefficient corresponding to each type of set pavement disease classification, when the radar imaging image of the asphalt pavement is classified and detected, the estimated confidence coefficient corresponding to the plurality of set pavement disease classifications can be determined only based on one radar imaging image detection network, and the disease detection is not carried out on each type of set pavement disease classification based on different radar imaging image detection networks.

From outside, based on the mechanism of debugging of two links, debug radar imaging image detection network, first link is based on first bituminous paving radar imaging image sample and the candidate road surface disease kind information of corresponding sample debug radar imaging image detection network, second link is based on second bituminous paving radar imaging image sample and the candidate road surface disease kind information of corresponding sample and carry out the secondary debugging to radar imaging image detection network, accomplish the fine debugging to radar imaging image detection network, help increasing radar imaging image detection network's precision.

In the embodiment of the present application, if the above-mentioned method for detecting the hidden diseases of the asphalt pavement is implemented in the form of a software functional module, and sold or used as a separate product, the method may also be stored in a computer readable storage medium. Based on such understanding, the technical solution of the embodiments of the present application may be essentially or some of contributing to the related art may be embodied in the form of a software product stored in a storage medium, including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read Only Memory (ROM), a magnetic disk, an optical disk, or other various media capable of storing program codes. Thus, embodiments of the application are not limited to any specific hardware, software, or firmware, or any combination of hardware, software, and firmware.

The embodiment of the application provides a computer device, which comprises a memory and a processor, wherein the memory stores a computer program capable of running on the processor, and the processor realizes part or all of the steps in the method when executing the program.

Embodiments of the present application provide a computer readable storage medium having stored thereon a computer program which, when executed by a processor, performs some or all of the steps of the above-described method. The computer readable storage medium may be transitory or non-transitory.

Embodiments of the present application provide a computer program comprising computer readable code which, when run in a computer device, causes a processor in the computer device to perform some or all of the steps for carrying out the above method.

Embodiments of the present application provide a computer program product comprising a non-transitory computer-readable storage medium storing a computer program which, when read and executed by a computer, performs some or all of the steps of the above-described method. The computer program product may be realized in particular by means of hardware, software or a combination thereof. In some embodiments, the computer program product is embodied as a computer storage medium, and in other embodiments, the computer program product is embodied as a software product, such as a software development kit (Software Development Kit, SDK), or the like.

It should be noted here that: the above description of various embodiments is intended to emphasize the differences between the various embodiments, the same or similar features being referred to each other. The above description of apparatus, storage medium, computer program and computer program product embodiments is similar to that of method embodiments described above, with similar advantageous effects as the method embodiments. For technical details not disclosed in the embodiments of the apparatus, the storage medium, the computer program and the computer program product of the present application, reference should be made to the description of the embodiments of the method of the present application.

Fig. 2 is a schematic diagram of a hardware entity of a computer device according to an embodiment of the present application, as shown in fig. 2, the hardware entity of the computer device 1000 includes: a processor 1001 and a memory 1002, wherein the memory 1002 stores a computer program executable on the processor 1001, the processor 1001 implementing the steps in the method of any of the embodiments described above when the program is executed.

The memory 1002 stores computer programs executable on the processor, the memory 1002 being configured to store instructions and applications executable by the processor 1001, and may also cache data (e.g., image data, audio data, voice communication data, and video communication data) to be processed or already processed by the respective modules in the processor 1001 and the computer device 1000, which may be implemented by a FLASH memory (FLASH) or a random access memory (Random Access Memory, RAM).

The processor 1001 performs the steps of the asphalt pavement hidden trouble detection method according to any one of the above. The processor 1001 generally controls the overall operation of the computer device 1000.

An embodiment of the present application provides a computer storage medium storing one or more programs executable by one or more processors to implement the steps of the asphalt pavement hidden disease detection method of any of the embodiments above.

It should be noted here that: the description of the storage medium and apparatus embodiments above is similar to that of the method embodiments described above, with similar benefits as the method embodiments. For technical details not disclosed in the embodiments of the storage medium and the apparatus of the present application, please refer to the description of the method embodiments of the present application. It will be appreciated that the electronic device implementing the above-mentioned processor function may be other, and embodiments of the present application are not limited in detail.

It should be appreciated that reference throughout this specification to "one embodiment" or "an embodiment" means that a particular feature, structure or characteristic described in connection with the embodiment is included in at least one embodiment of the present application. Thus, the appearances of the phrases "in one embodiment" or "in an embodiment" in various places throughout this specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. It should be understood that, in various embodiments of the present application, the sequence number of each step/process described above does not mean that the execution sequence of each step/process should be determined by its functions and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. The foregoing embodiment numbers of the present application are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments. It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.

In the several embodiments provided by the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. The above described device embodiments are only illustrative, e.g. the division of the units is only one logical function division, and there may be other divisions in practice, such as: multiple units or components may be combined or may be integrated into another system, or some features may be omitted, or not performed. In addition, the various components shown or discussed may be coupled or directly coupled or communicatively coupled to each other via some interface, whether indirectly coupled or communicatively coupled to devices or units, whether electrically, mechanically, or otherwise.

The units described above as separate components may or may not be physically separate, and components shown as units may or may not be physical units; can be located in one place or distributed to a plurality of network units; some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present application may be integrated in one processing unit, or each unit may be separately used as one unit, or two or more units may be integrated in one unit; the integrated units may be implemented in hardware or in hardware plus software functional units.

Those of ordinary skill in the art will appreciate that: all or part of the steps for implementing the above method embodiments may be implemented by hardware related to program instructions, and the foregoing program may be stored in a computer readable storage medium, where the program, when executed, performs steps including the above method embodiments; and the aforementioned storage medium includes: a mobile storage device, a Read Only Memory (ROM), a magnetic disk or an optical disk, or the like, which can store program codes.

Or the above-described integrated units of the application may be stored in a computer-readable storage medium if implemented in the form of software functional modules and sold or used as separate products. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the related art in the form of a software product stored in a storage medium, including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: various media capable of storing program codes, such as a removable storage device, a ROM, a magnetic disk, or an optical disk.

The foregoing is merely an embodiment of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily think about changes or substitutions within the technical scope of the present application, and the changes and substitutions are intended to be covered by the scope of the present application.

Claims

1. The method for detecting the hidden diseases of the asphalt pavement is characterized by comprising the following steps of:

Acquiring an asphalt pavement radar imaging image to be subjected to disease detection and a plurality of pieces of reference detection information corresponding to the asphalt pavement radar imaging image, wherein the plurality of pieces of reference detection information comprise confirmed past reference detection information corresponding to the asphalt pavement radar imaging image;

According to the set pavement disease classification corresponding to the multiple reference detection information, determining candidate pavement disease type information of the asphalt pavement radar imaging image; the candidate pavement disease type information comprises candidate confidence degrees of the asphalt pavement radar imaging images corresponding to multiple types of set pavement disease classifications;

Acquiring an integrated characterization vector of the asphalt pavement radar imaging image according to the asphalt pavement radar imaging image and the candidate pavement disease type information;

Performing disease detection on the asphalt pavement radar imaging image according to the integrated characterization vector to obtain a predicted disease detection result of the asphalt pavement radar imaging image; the estimated disease detection result comprises estimated confidence degrees of the asphalt pavement radar imaging image corresponding to the plurality of set pavement disease classifications, and the estimated disease detection result is used for determining the set pavement disease classifications corresponding to the asphalt pavement radar imaging image.

2. The method according to claim 1, wherein the determining candidate road surface disease type information of the asphalt road surface radar imaging image according to the set road surface disease classifications corresponding to the plurality of reference detection information includes:

Acquiring the number Q1 of each set pavement disease classification, wherein the number Q1 is the number of past reference detection information which belongs to the set pavement disease classification and is confirmed in the plurality of reference detection information;

And taking a first quotient result between the number Q1 of the set road surface disease classifications of each type and the total number of the plurality of reference detection information as a candidate confidence coefficient corresponding to the set road surface disease classifications of each type.

3. The method of claim 1, wherein the plurality of reference detection information further comprises repudiated past reference detection information corresponding to the asphalt pavement radar imaging image; the determining candidate pavement disease type information of the asphalt pavement radar imaging image according to the set pavement disease classification corresponding to the plurality of reference detection information comprises the following steps:

acquiring the number Q2 of each set pavement disease classification, wherein the number Q2 is the number of past reference detection information belonging to the set pavement disease classification in the plurality of reference detection information; obtaining a second quotient result between the number Q2 of each set pavement disease classification and the total number of the plurality of reference detection information;

Acquiring the number Q3 of each set pavement disease classification, wherein the number Q3 is the number of past reference detection information which belongs to the set pavement disease classification and is confirmed in the plurality of reference detection information; obtaining a third quotient result between the number Q3 of each set pavement disease classification and the total number of confirmed past reference detection information;

And respectively carrying out weight average on the second and third provider results of each class of set pavement disease classification to obtain candidate confidence degrees corresponding to each class of set pavement disease classification.

4. The method according to claim 1, wherein the performing disease detection on the asphalt pavement radar imaging image according to the integrated characterization vector to obtain a predicted disease detection result of the asphalt pavement radar imaging image comprises:

Extracting projection characterization vectors of a plurality of channels according to the integration characterization vectors;

According to the integrated characterization vector, obtaining an influence coefficient set of each type of set pavement disease classification, wherein the influence coefficient set comprises influence coefficients of the channels, and the influence coefficients represent the contribution degree of characterization information of an image on the channel to whether the image is the set pavement disease classification;

Respectively carrying out weight average on projection characterization vectors of the channels according to a plurality of influence coefficients in the influence coefficient set of each type of set pavement disease classification to obtain a type characterization vector of each type of set pavement disease classification;

And respectively carrying out disease detection on the asphalt pavement radar imaging image according to the type characterization vector of each type of set pavement disease classification to obtain the estimated confidence corresponding to each type of set pavement disease classification.

5. The method according to claim 1, wherein after performing disease detection on the asphalt pavement radar imaging image according to the integrated characterization vector to obtain a predicted disease detection result of the asphalt pavement radar imaging image, the method further comprises:

Taking the set pavement disease classification with the maximum estimated confidence as the set pavement disease classification corresponding to the asphalt pavement radar imaging image;

Determining a correlation coefficient between the asphalt pavement radar imaging image and each piece of undetermined reference detection information according to the set pavement disease classification corresponding to the asphalt pavement radar imaging image and the set pavement disease classification corresponding to the plurality of undetermined reference detection information corresponding to the asphalt pavement radar imaging image;

Taking undetermined reference detection information with the association coefficient meeting the association coefficient requirement as target detection information determined according to the asphalt pavement radar imaging image;

Before determining the candidate pavement disease type information of the asphalt pavement radar imaging image according to the set pavement disease classifications corresponding to the plurality of reference detection information, the method further comprises:

acquiring a plurality of road surface disease classifications corresponding to pre-stored undetermined reference detection information;

Acquiring the number Q4 corresponding to each road surface disease classification, wherein the number Q4 is the number of the pre-stored undetermined reference detection information belonging to the road surface disease classification;

And taking a plurality of road surface defect classifications, the number Q4 of which meets a number threshold value, as the plurality of set road surface defect classifications.

6. The method according to any one of claims 1 to 5, wherein the radar imaging image detection network includes a characterization vector extraction component and a disease detection component, the obtaining an integrated characterization vector of the asphalt pavement radar imaging image according to the asphalt pavement radar imaging image and the candidate pavement disease type information includes:

The integrated characterization vector is obtained according to the asphalt pavement radar imaging image and the candidate pavement disease type information through the characterization vector extraction component;

the disease detection is carried out on the asphalt pavement radar imaging image according to the integrated characterization vector to obtain a predicted disease detection result of the asphalt pavement radar imaging image, and the method comprises the following steps:

And performing disease detection on the asphalt pavement radar imaging image according to the integrated characterization vector by the disease detection assembly to obtain the estimated disease detection result.

7. The method according to claim 6, wherein the radar imaging image detection network is commissioned by:

Acquiring a first asphalt pavement radar imaging image sample and a plurality of sample reference detection information corresponding to the first asphalt pavement radar imaging image sample, wherein the sample reference detection information comprises confirmed past reference detection information corresponding to the first asphalt pavement radar imaging image sample;

Determining first candidate pavement disease type information of the first asphalt pavement radar imaging image sample according to set pavement disease classifications corresponding to the multiple sample reference detection information, wherein the first candidate pavement disease type information comprises sample candidate confidence coefficients of the first asphalt pavement radar imaging image sample corresponding to the multiple set pavement disease classifications;

Obtaining a sample integration characterization vector of the first asphalt pavement radar imaging image sample according to the first asphalt pavement radar imaging image sample and the first candidate pavement disease type information through the characterization vector extraction component;

performing disease detection on the first asphalt pavement radar imaging image sample according to the sample integration characterization vector through the disease detection assembly to obtain a first estimated disease detection result, wherein the first estimated disease detection result comprises sample estimated confidence degrees of the first asphalt pavement radar imaging image sample corresponding to the plurality of set pavement disease classifications;

According to a first priori disease detection result and the first estimated disease detection result of the first asphalt pavement radar imaging image sample, the radar imaging image detection network is debugged, so that errors between the first estimated disease detection result and the first priori disease detection result obtained by the radar imaging image detection network after debugging are reduced, and the first priori disease detection result comprises actual confidence that the first asphalt pavement radar imaging image sample corresponds to the classification of the various set pavement diseases.

8. The method of claim 7, wherein the token vector extraction component includes a first extraction unit, a second extraction unit, and a token information interaction unit, and wherein the obtaining, by the token vector extraction component, the sample integrated token vector for the first asphalt pavement radar imaging image sample according to the first asphalt pavement radar imaging image sample and the first candidate pavement disease type information includes:

Extracting, by the first extracting unit, a sample image characterization vector of the first asphalt pavement radar imaging image sample;

Extracting sample disease information characterization vectors of the first candidate pavement disease type information through the second extraction unit;

The sample image characterization vector and the sample disease information characterization vector are subjected to interactive integration through the characterization information interaction unit to obtain the sample integration characterization vector;

The disease detection assembly comprises a characterization vector projection unit corresponding to a plurality of channels, a weight adjustment unit for setting road surface disease classification of each class and a disease identification unit; the disease detection module is used for performing disease detection on the first asphalt pavement radar imaging image sample according to the sample integration characterization vector to obtain a first estimated disease detection result, and the method comprises the following steps:

Respectively integrating the characterization vectors according to the samples by a plurality of characterization vector projection units, and extracting sample projection characterization vectors of the channels;

the method comprises the steps that a weight adjusting unit of each type of set pavement disease classification is used for integrating a characterization vector according to the sample, and an influence coefficient set of each type of set pavement disease classification is obtained, wherein the influence coefficient set comprises influence coefficients of a plurality of channels, and the influence coefficients represent the contribution degree of characterization information of an image on the channels to whether the image is the set pavement disease classification or not;

respectively carrying out weight average on sample projection characterization vectors of the channels according to a plurality of influence coefficients in an influence coefficient set of each type of set pavement disease classification by a weight adjusting unit of each type of set pavement disease classification to obtain sample type characterization vectors of each type of set pavement disease classification;

And respectively carrying out disease detection on the first asphalt pavement radar imaging image samples according to the sample type characterization vectors of each type of set pavement disease classification by the disease identification unit of each type of set pavement disease classification to obtain the sample prediction confidence corresponding to each type of set pavement disease classification.

9. The method of claim 7, wherein the method further comprises, after commissioning the radar imaging image detection network based on the first prior disease detection result and the first predicted disease detection result of the first asphalt pavement radar imaging image sample:

acquiring a second asphalt pavement radar imaging image sample, and taking set pavement disease classification information as second candidate pavement disease type information of the second asphalt pavement radar imaging image sample, wherein the second candidate pavement disease type information comprises sample candidate confidence coefficients of the second asphalt pavement radar imaging image sample corresponding to the plurality of set pavement disease classifications;

Obtaining a sample integration characterization vector of the second asphalt pavement radar imaging image sample according to the second asphalt pavement radar imaging image sample and the second candidate pavement disease type information through the characterization vector extraction component; performing disease detection on the second asphalt pavement radar imaging image sample according to the sample integration characterization vector through the disease detection assembly to obtain a second estimated disease detection result, wherein the second estimated disease detection result comprises sample estimated confidence degrees of the second asphalt pavement radar imaging image sample corresponding to the plurality of set pavement disease classifications;

According to a second priori disease detection result and the second estimated disease detection result of the second asphalt pavement radar imaging image sample, the radar imaging image detection network is debugged, so that errors between the second estimated disease detection result obtained by the radar imaging image detection network after debugging and the second priori disease detection result are reduced, and the second priori disease detection result comprises actual confidence that the second asphalt pavement radar imaging image sample corresponds to the classification of the various set pavement diseases.

10. The method of claim 7, wherein the debugging the radar imaging image detection network according to the first priori disease detection result and the first estimated disease detection result of the first asphalt pavement radar imaging image sample, so that an error between the first estimated disease detection result and the first priori disease detection result obtained by the radar imaging image detection network after the debugging is reduced, comprises:

According to cost influence coefficients corresponding to each set pavement disease classification, carrying out weight average on cross entropy between the estimated confidence coefficient and the actual confidence coefficient of the sample of each set pavement disease classification to obtain a cost parameter, wherein the degree of commonality between the estimated confidence coefficient and the actual confidence coefficient of the sample is inversely related to the cross entropy;

And according to the cost parameter, debugging the radar imaging image detection network, and adjusting the cost influence coefficient, so that the cost parameter obtained according to the radar imaging image detection network after the debugging and the adjusted cost influence coefficient is reduced.

11. A computer device comprising a memory and a processor, the memory storing a computer program executable on the processor, characterized in that the processor implements the steps of the method of any of claims 1 to 10 when the program is executed.