CN117851778B - Apparent recognition method and system based on bridge exposure degree - Google Patents

Apparent recognition method and system based on bridge exposure degree Download PDF

Info

Publication number
CN117851778B
CN117851778B CN202410030781.2A CN202410030781A CN117851778B CN 117851778 B CN117851778 B CN 117851778B CN 202410030781 A CN202410030781 A CN 202410030781A CN 117851778 B CN117851778 B CN 117851778B
Authority
CN
China
Prior art keywords
disease
bridge
frequency
area
type
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202410030781.2A
Other languages
Chinese (zh)
Other versions
CN117851778A (en
Inventor
刘天成
朱超
王伟
沈大为
王小宁
程潜
鲜荣
郭国和
王杨
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangdong Provincial Highway Construction Co ltd
CCCC Highway Long Bridge Construction National Engineering Research Center Co Ltd
Original Assignee
Guangdong Provincial Highway Construction Co ltd
CCCC Highway Long Bridge Construction National Engineering Research Center Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guangdong Provincial Highway Construction Co ltd, CCCC Highway Long Bridge Construction National Engineering Research Center Co Ltd filed Critical Guangdong Provincial Highway Construction Co ltd
Priority to CN202410030781.2A priority Critical patent/CN117851778B/en
Publication of CN117851778A publication Critical patent/CN117851778A/en
Application granted granted Critical
Publication of CN117851778B publication Critical patent/CN117851778B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/08—Construction
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Business, Economics & Management (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Human Resources & Organizations (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Economics (AREA)
  • Primary Health Care (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • Image Analysis (AREA)

Abstract

The invention provides an apparent recognition method and system based on bridge exposure, which relate to the technical field of data processing and comprise the following steps: loading bridge foundation information, including bridge structure topology, deployment climate characteristics and deployment geographic characteristics; performing disease statistical analysis to generate the disease exposure degree and the high-frequency disease type of the high-frequency disease area; configuring an apparent disease recognition main model of a high-frequency disease area based on the high-frequency disease type with the disease exposure degree larger than or equal to the disease exposure degree threshold value; deleting the high-frequency disease type from the first disease type set to obtain a second disease type set; constructing an apparent disease identification auxiliary model by combining a twin neural network; and merging and constructing an apparent disease identification component to process the bridge monitoring image, generating a bridge disease identification area and sending the bridge disease identification area to a user side. The method solves the technical problems of poor accuracy and low recognition efficiency of the traditional bridge disease recognition method.

Description

Apparent recognition method and system based on bridge exposure degree
Technical Field
The invention relates to the technical field of data processing, in particular to an apparent recognition method and system based on bridge exposure.
Background
The bridge is an important component of the infrastructure, and various apparent diseases such as cracks, flaking, exposed ribs, honeycomb pitting and the like can be generated in the long-term use process of the cross-sea bridge due to the reasons of increased traffic flow, overload of vehicles, structural aging and the like. Because of the gradual increase of diseases and complex bridge structure, great inconvenience is brought to the conventional disease detection mainly based on manual detection, and a new disease identification and detection technology is needed to be developed.
In the period of low development level of various automatic tools and image processing technologies, the identification and detection work of the apparent bridge diseases is mainly carried out manually, and bridge inspection staff needs to periodically detect and record the diseases of each part of the bridge. Because of randomness and universality of bridge apparent diseases, a large amount of manpower is consumed for manually detecting and recording the bridge apparent diseases, the detection speed is low, and a large potential safety hazard exists in the practical detection process. In addition, since subjective judgment of a recorder may affect accuracy of a measurement result in the recording process, a purely manual method has high requirements on experience and capability of a measurer. Due to the differences in habits and experience, the results measured by different measurement personnel deviate greatly and the referenceability of the measurement results may be reduced. The traditional bridge disease detection mode using manual detection as a main body gradually fails to meet the requirements of low labor, low cost and high efficiency of the current bridge disease identification detection.
Disclosure of Invention
The application provides an apparent recognition method based on bridge exposure, which aims to solve the technical problems that various characteristics of a bridge cannot be comprehensively considered in the traditional bridge disease recognition method, and the accuracy and the recognition efficiency of disease recognition are low due to the lack of a large-scale bridge monitoring data processing and analyzing method.
In view of the above problems, the application provides an apparent recognition method and system based on bridge exposure.
In a first aspect of the disclosure, an apparent recognition method based on bridge exposure is provided, the method comprising: loading bridge foundation information, wherein the bridge foundation information comprises bridge structure topology, deployment climate characteristics and deployment geographic characteristics; performing disease statistical analysis by taking the bridge structure topology, the deployment climate characteristics and the deployment geographic characteristics as constraints, and generating the disease exposure and the high-frequency disease type of a high-frequency disease area; configuring an apparent disease recognition master model of the high-frequency disease region based on the high-frequency disease type in which the disease exposure is greater than or equal to a disease exposure threshold; deleting the high-frequency disease type from the first disease type set to obtain a second disease type set; based on the second disease type set, combining a twin neural network to build an apparent disease identification auxiliary model; and combining the apparent disease identification main model and the apparent disease identification auxiliary model, constructing an apparent disease identification assembly, processing a bridge monitoring image, generating a bridge disease identification area and sending the bridge disease identification area to a user side.
In another aspect of the disclosure, there is provided an apparent recognition system based on bridge exposure, the system being used in the above method, the system comprising: the system comprises a basic information loading module, a bridge information processing module and a bridge information processing module, wherein the basic information loading module is used for loading bridge basic information, and the bridge basic information comprises bridge structure topology, deployment climate characteristics and deployment geographic characteristics; the disease statistics analysis module is used for carrying out disease statistics analysis by taking the bridge structure topology, the deployment climate characteristics and the deployment geographic characteristics as constraints to generate the disease exposure degree and the high-frequency disease type of a high-frequency disease area; the main model configuration module is used for configuring an apparent disease identification main model of the high-frequency disease area based on the high-frequency disease type with the disease exposure degree larger than or equal to a disease exposure degree threshold value; the high-frequency disease deleting module is used for deleting the high-frequency disease type from the first disease type set to obtain a second disease type set; the auxiliary model configuration module is used for constructing an apparent disease identification auxiliary model based on the second disease type set and combining a twin neural network; the identification region generation module is used for combining the apparent disease identification main model and the apparent disease identification auxiliary model, constructing an apparent disease identification component, processing a bridge monitoring image, generating a bridge disease identification region and sending the bridge disease identification region to a user side.
One or more technical schemes provided by the application have at least the following technical effects or advantages:
By loading bridge foundation information, including structural topology, climate characteristics and geographic characteristics, various characteristics of the bridge can be comprehensively considered, and comprehensive constraint conditions are provided for subsequent disease analysis; through disease statistical analysis, the disease exposure degree and the high-frequency disease type of a high-frequency disease area are generated, so that the system can focus on the problem with the most obvious influence on the bridge structure; the high-frequency disease type is used as constraint, the apparent disease identification main model and the auxiliary model are configured, and the twin neural network is combined to effectively identify and classify the apparent diseases of the bridge, so that the accuracy of disease detection is improved; the apparent disease identification component is constructed by combining the main model and the auxiliary model, so that the comprehensive processing of bridge monitoring images is realized, and a bridge disease identification area is generated, thereby providing timely and accurate disease information for a user side. In conclusion, the method effectively solves the problems of the traditional bridge disease monitoring method by integrating bridge foundation information and adopting the disease statistical analysis and apparent disease identification model, and improves the accuracy and efficiency of bridge diseases.
The foregoing description is only an overview of the present application, and is intended to be implemented in accordance with the teachings of the present application in order that the same may be more clearly understood and to make the same and other objects, features and advantages of the present application more readily apparent.
Drawings
FIG. 1 is a schematic flow chart of an apparent recognition method based on bridge exposure according to an embodiment of the application;
fig. 2 is a schematic structural diagram of an apparent recognition system based on bridge exposure according to an embodiment of the present application.
Reference numerals illustrate: the system comprises a basic information loading module 10, a disease statistics analysis module 20, a main model configuration module 30, a high-frequency disease deletion module 40, an auxiliary model configuration module 50 and an identification area generation module 60.
Detailed Description
The embodiment of the application solves the technical problems that various characteristics of a bridge cannot be comprehensively considered in the traditional bridge disease identification method and the accuracy and the identification efficiency of disease identification are low due to the lack of a large-scale bridge monitoring data processing and analyzing method by providing the apparent identification method based on the bridge exposure.
Having described the basic principles of the present application, various non-limiting embodiments of the present application will now be described in detail with reference to the accompanying drawings.
Example 1
As shown in fig. 1, an embodiment of the present application provides an apparent recognition method based on bridge exposure, where the method includes:
loading bridge foundation information, wherein the bridge foundation information comprises bridge structure topology, deployment climate characteristics and deployment geographic characteristics;
The bridge structure topology describes the construction and connection modes of the bridge, which are critical to the propagation and influence of diseases, the structure topology comprises information of various components, supporting structures and the like of the bridge, and CAD (computer aided design) software or other structure design tools are utilized to extract the structure topology of the bridge.
The deployment climate characteristics comprise data related to the climate of the area where the bridge is located, such as temperature, humidity, rainfall and the like, the characteristics have an influence on the occurrence and evolution of diseases, real-time or historical meteorological data are acquired through a nearby meteorological station, or meteorological characteristics of the area are simulated and predicted by using a meteorological model, and the deployment climate characteristics are acquired.
The deployment geographic features comprise information such as the topography, the height, the water distribution and the like of the bridge location, the information also has an influence on the formation and the propagation of diseases, geographic information around the bridge is obtained by utilizing a GIS tool, the information comprises the topography, the water distribution and the like, and the laser radar technology is used for obtaining the topography and the height information.
And integrating the collected bridge structure topology, deployment climate characteristics and deployment geographic characteristics to obtain bridge foundation information, and laying a foundation for subsequent disease statistical analysis.
Performing disease statistical analysis by taking the bridge structure topology, the deployment climate characteristics and the deployment geographic characteristics as constraints, and generating the disease exposure and the high-frequency disease type of a high-frequency disease area;
Further, performing disease statistical analysis by taking the bridge structure topology, the deployment climate characteristics and the deployment geographical characteristics as constraints, and generating the disease exposure and the high-frequency disease type of the high-frequency disease area, wherein the method comprises the following steps:
Taking the bridge structure topology, the deployment climate characteristics and the deployment geographic characteristics as constraints, and collecting bridge disease record data based on a bridge cloud database, wherein any one piece of the bridge disease record data comprises a bridge disease record position and a bridge disease record type;
based on the bridge defect recording type, performing position expansion on the bridge defect recording position in the bridge structure topology to generate a bridge defect recording area;
Clustering the bridge defect recording areas based on an area intersection area ratio threshold value to generate a bridge defect recording area clustering result;
Sorting based on the clustering result of the bridge disease recording area to generate the high-frequency disease area and the high-frequency disease type, wherein the high-frequency disease area and the high-frequency disease type are in one-to-one correspondence;
and carrying out exposure degree analysis on the high-frequency disease area to generate the disease exposure degree, wherein the disease exposure degree is equal to the area high-frequency factor of the high-frequency disease area divided by an area high-frequency factor threshold value.
The system is connected to a bridge cloud database, so that the system can access information in the database, and a query request is sent to the database by using the information such as structural topology, climate characteristics, geographical characteristics and the like as constraints to acquire bridge defect record data related to the characteristics, wherein the bridge defect record type can be various possible defects such as cracks, flaking, exposed ribs and the like.
For the type of fault recording, the fault recording locations are expanded in the structural topology of the bridge, in particular for corrosive faults, structural elements around the expanded recording locations can be considered, as corrosion can lead to local damage of the structural elements; for non-corrosive diseases, different propagation modes are used depending on the specific type, for example, for cracks, it is conceivable to perform a region propagation around the crack location.
And generating bridge disease recording areas in the bridge structure topology according to the information after the position expansion, wherein the areas represent specific positions possibly affected by diseases, and storing the expanded position information and corresponding disease recording types in a correlated manner so as to facilitate subsequent disease statistical analysis and model training.
And (3) comparing all bridge disease recording areas in pairs to calculate an intersection area ratio, wherein the specific calculation method is to calculate the sum of the non-intersection area and the intersection area of the two areas, and then calculate the sum of the two areas to obtain the intersection area ratio. Judging whether the intersection area ratio is larger than or equal to the intersection area ratio threshold value of the regions, and when the intersection area ratio meets the condition, gathering the two regions into one type, which means that the two regions possibly belong to the same disease type or are affected by the similarity. All clustered disease recording regions are arranged into bridge disease recording region clustering results, each class in the clustering results may contain a plurality of disease recording regions, and the regions may share similar disease characteristics.
Traversing the bridge disease record region clustering result, obtaining a region list of each type, and counting the number of regions of each type to obtain a region high-frequency factor, namely the number of regions in the type; traversing the clustering result of the bridge disease record area, obtaining a disease record type list of each type, and counting the occurrence frequency of each disease record type in each type to obtain a disease high-frequency factor, namely the occurrence frequency of each disease type in the type.
And fusing the high-frequency areas according to the set area high-frequency factor threshold, for example, polymerizing the areas with the area high-frequency factors larger than or equal to the threshold into a high-frequency disease area, and determining the high-frequency disease type according to the disease type with the generated high-frequency disease area with the disease high-frequency factors larger than or equal to the disease high-frequency factor threshold.
And acquiring the area high-frequency factor of each high-frequency disease area, wherein the area high-frequency factor refers to the frequency parameter triggered by the disease area, namely the frequency of the area in the clustering result. And setting a region high-frequency factor threshold value, wherein the threshold value is used for judging which high-frequency disease regions need exposure analysis. For each high-frequency disease region, calculating the disease exposure, namely calculating the ratio of the region high-frequency factor to the region high-frequency factor threshold, and correlating the calculated disease exposure with the corresponding high-frequency disease region to form a final disease exposure result.
The exposure degree analysis is carried out by utilizing the regional high-frequency factors of the high-frequency disease regions, and the frequent occurrence condition of the high-frequency disease regions is measured by calculating the disease exposure degree, so that the method is favorable for further understanding the possible high-frequency diseases in the bridge structure and provides important information for subsequent apparent disease identification.
Further, based on the bridge defect recording type, performing position expansion on the bridge defect recording position in the bridge structure topology to generate a bridge defect recording area, including:
Dividing the bridge disease record type into a corrosive disease type and a non-corrosive disease type;
performing position expansion based on the bridge disease recording position according to the region expansion radius to generate a first bridge disease recording region, wherein the first bridge disease recording region and the corrosion disease type are stored in an associated mode;
Performing position expansion based on the bridge defect recording position according to the region expansion radius and the element splicing boundary to generate a second bridge defect recording region, wherein the second bridge defect recording region and the non-corrosive type are stored in an associated mode;
And adding the second bridge defect recording area and the first bridge defect recording area into the bridge defect recording area.
And judging each bridge disease record according to the disease type, and classifying the bridge disease record into a corrosion disease type or a non-corrosion disease type, wherein if the disease type belongs to the corrosion disease, the bridge disease record is classified into the corrosion disease type, and if the disease type belongs to the non-corrosion disease, the bridge disease record is classified into the non-corrosion disease type. And storing each divided bridge disease record in a correlated way with the corrosion disease type or the non-corrosion disease type to which the bridge disease record belongs.
Setting a radius for expanding the bridge defect record position, wherein the radius is used for determining an expansion range, traversing all bridge defect records divided into corrosive defect types, carrying out position expansion on each bridge defect record of the corrosive defect types by using the set expansion radius as a center, generating a first bridge defect record area, and storing the generated first bridge defect record area and the corresponding corrosive defect type in an associated manner, so that each area is associated with the corresponding corrosive defect type.
For non-corrosive types of lesions, considering the characteristic that the bridge is assembled by splicing multiple modules, the element splice boundaries are set, i.e., expansion is stopped when a boundary or radius endpoint is encountered in each direction. Traversing all bridge defect records divided into non-corrosive defect types, and performing position expansion on each non-corrosive defect type bridge defect record by using the position of each non-corrosive defect type bridge defect record as a center and using a set expansion radius and element splicing boundaries to generate a second bridge defect record area. And storing the generated second bridge disease record area in association with the corresponding non-corrosive disease type, so that each area is associated with the corresponding non-corrosive disease type.
And combining the first bridge defect recording area and the second bridge defect recording area to generate an integral set of bridge defect recording areas, and continuously maintaining the association storage of the combined bridge defect recording areas with corresponding corrosive defect types or non-corrosive defect types, so that two recording areas can be possibly expanded in one recording position, and each recording area is provided with a group of defect type identifiers.
Further, based on a threshold value of the area-to-area ratio of the intersection of the areas, the bridge defect recording areas are clustered, and a clustering result of the bridge defect recording areas is generated, including:
Performing intersection analysis on the bridge disease recording areas in pairs to generate area intersection areas and area non-intersection areas which are in one-to-one correspondence;
calculating the sum of the area intersection area and the area non-intersection area to generate an area summation area;
calculating the ratio of the area of the intersection of the areas to the sum area of the areas, and setting the ratio as the area ratio of the intersection of the areas;
And aggregating the bridge defect recording areas with the area intersection area ratio being larger than or equal to the area intersection area ratio threshold value into one type, and generating a clustering result of the bridge defect recording areas.
Traversing all bridge disease recording areas, combining two by two to obtain the combination of each pair of areas, calculating the intersection area of each pair of bridge disease recording areas by calculating the area of the overlapping part of the two areas, calculating the intersection area of the two areas, and subtracting the intersection area to obtain the non-intersection area.
And calculating the sum of the intersection area and the non-intersection area of each pair of bridge disease recording areas to obtain an area addition area.
And calculating the ratio of the area intersection area to the area summation area of each pair of bridge disease recording areas to obtain the area intersection area ratio.
A threshold value of the intersection area ratio of the areas is preset to judge whether the two areas should be gathered into one type, and the threshold value can be set according to specific application scenes and requirements. And judging whether the area ratio of intersection areas of each pair of bridge disease recording areas is larger than or equal to a set threshold value, and if the conditions are met, aggregating the two areas into one type. Traversing all bridge defect recording area combinations, storing clustered results, and generating bridge defect recording area clustering results, wherein each type of bridge defect recording areas meet the conditions.
Further, sorting based on the clustering result of the bridge disease recording area, generating the high-frequency disease area and the high-frequency disease type, including:
Traversing the clustering result of the bridge disease recording area, and counting area high-frequency factors, wherein the area high-frequency factors are equal to the number of the areas in the class;
Traversing the clustering result of the bridge disease recording area, and counting the high-frequency factors of the diseases, wherein the high-frequency factors of the diseases are equal to the occurrence frequency of each disease type in the class, and the high-frequency factors of the area and the high-frequency factors of the diseases are in one-to-many association;
Performing region fusion on the clustering result of the bridge disease recording region with the region high-frequency factor larger than or equal to the region high-frequency factor threshold value to generate the high-frequency disease region;
setting the disease type of the high-frequency disease area, wherein the disease high-frequency factor of the high-frequency disease area is larger than or equal to a disease high-frequency factor threshold value, as the high-frequency disease type of the high-frequency disease area.
Traversing the clustering result of the bridge defect recording areas, accessing the bridge defect recording areas in each cluster, counting the number of the bridge defect recording areas, namely the number of the intra-class areas, for each cluster, taking the number of the intra-class areas in each cluster as the area high-frequency factors of the cluster, wherein the area high-frequency factors represent the occurrence frequency of the defects in the class, and the more the number is, the more common the defects in the class are indicated.
Traversing the clustering result of the bridge disease recording areas, accessing the bridge disease recording areas in each cluster, and counting the occurrence frequency of different disease types in each cluster, wherein the occurrence frequency of each disease type in the cluster can be used as a disease high-frequency factor of the disease type by analyzing the disease recording type of each area, and the higher the disease high-frequency factor shows the frequency of the disease type in a specific cluster, the more common the disease type in the cluster. And establishing a one-to-many association relation of the regional high frequency factors and the disease high frequency factors, namely, associating each regional high frequency factor to the disease high frequency factors of all disease types in the cluster.
And traversing the clustering results of the bridge disease recording areas, and only selecting the clustering results of which the area high frequency factor is larger than or equal to a preset area high frequency factor threshold value, so that only those disease areas with higher frequency in the corresponding clusters are selected. For the selected high-frequency disease areas, the area fusion operation is performed, and the area fusion can adopt some image processing and analysis technologies, such as image segmentation, area fusion and the like, wherein the aim is to combine adjacent high-frequency disease areas into larger continuous areas, and after the area fusion, the high-frequency disease areas are generated, wherein the areas are the disease areas with higher frequency in the whole bridge structure.
And traversing each disease type in the high-frequency disease area, and selecting only the disease types of which the disease high-frequency factors are greater than or equal to a preset disease high-frequency factor threshold value, so that only those disease types with higher frequencies in the high-frequency disease area are selected. The disease types that meet the conditions are set as high-frequency disease types in the high-frequency disease region, and these high-frequency disease types are usually disease types that occur with a high frequency in the region.
Configuring an apparent disease recognition master model of the high-frequency disease region based on the high-frequency disease type in which the disease exposure is greater than or equal to a disease exposure threshold;
further, configuring the apparent disease recognition master model of the high-frequency disease region based on the high-frequency disease type in which the disease exposure is greater than or equal to a disease exposure threshold, comprising:
the bridge foundation information also comprises appearance shape characteristics and appearance color characteristics;
traversing the high-frequency disease type to acquire a first bridge disease image data set based on the appearance shape characteristic and the appearance color characteristic, wherein the first bridge disease image data set comprises a first bridge disease original image data set and a first bridge disease identification image data set with a high-frequency disease type identification;
The first bridge disease identification image data set is provided with a first bridge disease type identification image data set, a second bridge disease type identification image data set and an Nth bridge disease type identification image data set;
The first bridge disease type identification image data set is used as supervision data, the first bridge disease original image data set is used as input data, and a first disease type identification channel is configured based on a convolutional neural network;
The N-th bridge disease type identification channel is configured based on a convolutional neural network by taking the N-th bridge disease type identification image data set as supervision data and the first bridge disease original image data set as input data;
And connecting the first disease type identification channel to the Nth disease type identification channel as parallel nodes, merging input layers, and generating the apparent disease identification main model.
Appearance shape features, including bridge geometry, structural contours, etc., are extracted from bridge monitoring images by image processing techniques or computer vision algorithms, such as edge detection, contour extraction, shape descriptors, etc. Also, appearance color features are extracted from the bridge monitoring image, including information such as color distribution, color gradient and the like of the bridge surface, and the color feature extraction involves conversion of a color space, such as from RGB to HSV and the like.
According to the appearance shape characteristics and the appearance color characteristics, traversing the determined high-frequency disease types, selecting samples with the disease types from the bridge monitoring images aiming at each high-frequency disease type, and dividing the samples into two types, wherein a first bridge disease original image data set comprises an unidentified original bridge disease image, and a first bridge disease identification image data set comprises a corresponding identification, namely a bridge disease image identified by the high-frequency disease type.
The acquired image data is organized into data sets, each sample including an appearance shape feature, an appearance color feature, and a corresponding disease type identifier. For example, assuming that three high frequency lesion types have been determined, including cracking, corrosion and vibration, for each lesion type, a sample with the corresponding lesion is selected from the monitored image, and for the crack type, including the original image and the identification image of the crack lesion, there will be corresponding data sets for the corrosion and vibration types as well.
This process ensures that the training dataset contains samples of various high frequency disease types, so that the apparent disease recognition master model can learn the characteristics of different disease types during the training phase.
For each high-frequency disease type, the collected image data sets with the disease marks are integrated into a first bridge disease mark image data set, each data set comprises mark image data of the corresponding high-frequency disease type, wherein the structure of the first bridge disease mark image data set is a set, each element is one disease type mark image data set, the data sets respectively correspond to the first bridge disease type and the second bridge disease type and up to the Nth bridge disease type, the diversity of the data sets is ensured, and the information of a plurality of high-frequency disease types can be covered. For example, for a crack type, there is a crack disease identification image dataset; for corrosion types, there is a corrosion disease identification image dataset; and so on, these datasets are combined into a first bridge disease identification image dataset.
The construction of the first bridge disease identification image data set ensures that the model can learn the identification characteristics of different high-frequency disease types in a training stage and has generalization capability for various diseases.
By using convolutional neural networks (Convolutional Neural Network, CNN, which are common architecture for image processing tasks in deep learning, spatial features of images can be effectively captured), a first lesion type identification channel is configured, specifically, a first bridge lesion type identification image dataset containing images with a first lesion type identification is used as supervisory data, and a first bridge lesion raw image dataset is used as input data, which is an unlabeled raw bridge lesion image.
Using a Convolutional Neural Network (CNN) as the architecture of the model, a first lesion type recognition channel is configured, the goal of which is to learn to extract features related to the first lesion type from the original bridge lesion image, this channel consisting of a convolutional layer, a pooling layer and a fully connected layer.
The method comprises the steps of using a first bridge disease type identification image dataset as a supervision signal, training a configured network through a back propagation algorithm and an optimization algorithm such as gradient descent, wherein the training aims at minimizing the gap between a prediction result and a real label so that the network can accurately identify a first disease type, and storing model parameters of a configured first disease type identification channel when training is completed so as to be used in subsequent disease identification.
Similar to the previous process of configuring the first disease type recognition channel, the second to nth disease type recognition channels are configured to learn the characteristics of the second to nth disease types, respectively, and are not described herein again for brevity of the description.
The first disease type identification channel is connected to the Nth disease type identification channel to form parallel nodes, which means that each disease type has an independent channel to learn the characteristics of the type, the parallel nodes are connected to a common input layer, the input layer receives bridge disease original image data as input, the characteristics of each disease type can be considered in a network, the parallel nodes and the input layer are integrated to form an integral network structure, each channel learns the characteristics of different disease types, and the input layer is responsible for receiving the original image, so that an apparent disease identification main model is generated, and the model can simultaneously consider the information of different disease types to identify the apparent disease in a more comprehensive mode.
Deleting the high-frequency disease type from the first disease type set to obtain a second disease type set;
The user predefines a set of disease types as a first set of disease types that contains all possible bridge disease types that may cover all diseases that can be detected by the bridge monitoring system. Deleting the obtained high-frequency disease types from the first disease type set preset by the user, and taking the obtained residual disease types as a second disease type set.
Based on the second disease type set, combining a twin neural network to build an apparent disease identification auxiliary model;
further, based on the second disease type set, combining with a twin neural network, constructing an apparent disease identification auxiliary model, including:
Based on the appearance shape features and the appearance color features, traversing the second disease type set to acquire a second bridge disease image dataset, wherein the second bridge disease image dataset comprises a second bridge disease original image dataset, a disease texture feature identification dataset and a color feature identification dataset;
The disease texture feature identification data set is used as supervision data, the second bridge disease original image data set is used as input data, and a texture feature extraction channel is configured based on a convolutional neural network;
The color feature identification data set is used as supervision data, the second bridge disease original image data set is used as input data, and a color feature extraction channel is configured based on a convolutional neural network;
the texture feature extraction channel and the color feature extraction channel are used as parallel nodes to be fully connected, so that a feature extraction channel is generated;
And (3) configuring a feature comparison function:
Wherein, P represents a feature comparison output value, p=1 represents success of comparison, p=0 represents failure of comparison, s c represents color similarity, s t represents texture similarity, s c0 represents a color similarity threshold, and s t0 represents a texture similarity threshold;
configuring a color similarity calculation channel and a texture similarity calculation channel;
Copying the feature extraction channels to construct a first feature extraction channel and a second feature extraction channel, wherein model parameters of the first feature extraction channel and the second feature extraction channel are identical to those of the feature extraction channels;
combining the first feature extraction channel and the second feature extraction channel as parallel channels, constructing an input processing layer, combining the color similarity calculation channel and the texture similarity calculation channel as parallel channels, constructing a comparison processing layer, and constructing an output processing layer based on the feature comparison function;
And sequentially connecting the input processing layer, the comparison processing layer and the output processing layer in series to generate the apparent disease identification auxiliary model.
Traversing the bridge data set for each disease type in the second disease type by using the appearance shape features and the appearance color features, and collecting image data related to the current disease type, wherein the image data comprises a second bridge disease original image data set, a disease texture feature identification data set and a color feature identification data set, and the second bridge disease original image data set comprises an unprocessed disease image; the lesion texture feature identification dataset contains identifications of texture features, possibly descriptions of lesion surface textures; the color characterization data set contains a characterization of the color characterization, possibly a description of the disease color. The acquired image data is organized into a second bridge disease image dataset, ensuring that each image is associated with its corresponding disease type identification.
Preparing training data, wherein the training data comprises a disease texture feature identification data set as supervision data and a second bridge disease original image data set as input data, constructing a model based on a Convolutional Neural Network (CNN), wherein the model comprises a channel which is specially used for extracting the texture features of the disease, namely a texture feature extraction channel, using the disease texture feature identification data set as supervision data, training the network to learn to extract the texture features from the disease original image, and adjusting network parameters through a back propagation algorithm and a gradient descent optimization algorithm so as to accurately identify the features related to the disease texture. And performing supervised learning by using the disease texture feature identification data set, training CNN, wherein the aim of the supervised learning is to enable a texture feature extraction channel to learn effective disease texture feature representation, iterating the training process until convergence, and acquiring the texture feature extraction channel after model convergence, wherein the convergence condition can be that the preset iteration times are reached or the preset accuracy is met.
The same method as the configuration of the texture feature extraction channel is adopted, the color feature identification data set is used as supervision data, the second bridge disease original image data set is used as input data, and the color feature extraction channel is configured based on a Convolutional Neural Network (CNN), so that the description is omitted here for brevity.
The texture feature extraction channel and the color feature extraction channel are connected as parallel nodes, which means that the two channels process input data at the same time to extract texture features and color features respectively, the output of the parallel nodes is integrated through a full connection layer, the full connection layer is used for effectively combining the features from the two channels together to form a final feature extraction channel, and the design is helpful for comprehensively considering different aspects of images and improving the accuracy of disease identification.
The feature comparison function is configured, the function receives the color similarity and the texture similarity as input, and a comparison result is output according to a preset color similarity threshold value and a preset texture similarity threshold value, wherein the threshold value is adjusted according to specific requirements and data characteristics so as to ensure that whether the similarity of the color and the texture reaches an expected level can be accurately judged in practical application by the comparison function. For example, if the color similarity threshold s c0＝0.8,st0 =0.7 is set, when the color similarity s c is equal to or greater than 0.8 and the texture similarity s t is equal to or greater than 0.7, the output P of the comparison function will be 1, indicating that the comparison is successful, and in other cases, the output P is 0, indicating that the comparison is failed.
Color similarity calculation and texture similarity calculation according to user requirements, an existing arbitrary calculation mode is configured, specifically, an image is converted from an RGB color space into other color spaces, such as Lab, HSV and the like, an appropriate color space is selected according to specific requirements, for each channel, for example, for the Lab space, three channels L, a and b are provided, color histograms are extracted, the color histograms represent distribution of different colors in the image, the color histograms in the selected color space are used, and the color similarity can be calculated by using different methods, such as histogram intersection, cosine similarity and the like; texture features are extracted from the image using texture feature extraction algorithms, such as Gabor filters, and using the extracted texture features, different methods may be used to calculate texture similarity, such as euclidean distance, correlation coefficients, and the like.
The model parameters of the configured feature extraction channels, including weights, offsets, etc., are replicated, and the two feature extraction channels are constructed in a parallel structure, ensuring that they accept the same input image at the same time.
And receiving the bridge monitoring image as input through the input layer, obtaining the first type of characteristics of the bridge image through the first characteristic extraction channel, obtaining the second type of characteristics of the bridge image through the second characteristic extraction channel, and connecting the first characteristic extraction channel and the second characteristic extraction channel as parallel channels to construct the input processing layer.
The color similarity calculating channel calculates color similarity, the texture similarity calculating channel calculates texture similarity, and the color similarity calculating channel and the texture similarity calculating channel are connected as parallel channels to construct a comparison processing layer.
A feature comparison function is used that accepts color similarity and texture similarity as inputs and outputs the result of success or failure of the comparison. And connecting the output of the input processing layer and the output of the comparison processing layer together, and using the output obtained by the characteristic comparison function as a final output result.
The design can effectively process the feature extraction and similarity comparison of bridge images, and generates a final output result through a feature comparison function to judge whether diseases exist.
The input processing layer, the comparison processing layer and the output processing layer are sequentially connected in series, so that an apparent disease identification auxiliary model is obtained, the auxiliary model can receive bridge monitoring images and judge whether diseases exist according to comparison results of color and texture similarity, and the design of the auxiliary model considers the overall flow of feature extraction, similarity calculation and comparison processing so as to effectively process bridge images and generate corresponding identification results.
And combining the apparent disease identification main model and the apparent disease identification auxiliary model, constructing an apparent disease identification assembly, processing a bridge monitoring image, generating a bridge disease identification area and sending the bridge disease identification area to a user side.
And integrating the apparent disease recognition main model and the auxiliary model into an integral apparent disease recognition assembly through modes of model fusion, stacking and the like, so that the main model and the auxiliary model can be ensured to work cooperatively.
The bridge monitoring image is input into an integrated apparent disease recognition component, and the apparent disease recognition component extracts appearance features, color features, texture features and the like of the image by utilizing the main model and the auxiliary model according to the bridge monitoring image. The image features subjected to feature extraction are input into an integrated apparent disease recognition component, and the component recognizes various possible diseases in the image according to the learned knowledge, and generates a bridge disease identification area, namely, the position of the disease in the image is determined, wherein the position can be in the form of a rectangular frame, a polygon and the like.
The generated bridge disease identification area information is transmitted to the user side through network transmission or other communication means, so that automatic identification and positioning of bridge diseases are realized.
In summary, the apparent recognition method and system based on bridge exposure provided by the embodiment of the application have the following technical effects:
1. By loading bridge foundation information, including structural topology, climate characteristics and geographic characteristics, various characteristics of the bridge can be comprehensively considered, and comprehensive constraint conditions are provided for subsequent disease analysis;
2. through disease statistical analysis, the disease exposure degree and the high-frequency disease type of a high-frequency disease area are generated, so that the system can focus on the problem with the most obvious influence on the bridge structure;
3. The high-frequency disease type is used as constraint, the apparent disease identification main model and the auxiliary model are configured, and the twin neural network is combined to effectively identify and classify the apparent diseases of the bridge, so that the accuracy of disease detection is improved;
4. the apparent disease identification component is constructed by combining the main model and the auxiliary model, so that the comprehensive processing of bridge monitoring images is realized, and a bridge disease identification area is generated, thereby providing timely and accurate disease information for a user side.
In conclusion, the method effectively solves the problems of the traditional bridge disease monitoring method by integrating bridge foundation information and adopting the disease statistical analysis and apparent disease identification model, and improves the accuracy and efficiency of bridge diseases.
Example two
Based on the same inventive concept as the apparent recognition method based on bridge exposure in the foregoing embodiments, as shown in fig. 2, the present application provides an apparent recognition system based on bridge exposure, the system comprising:
The system comprises a basic information loading module 10, wherein the basic information loading module 10 is used for loading bridge basic information, and the bridge basic information comprises bridge structure topology, deployment climate characteristics and deployment geographic characteristics;
The disease statistics analysis module 20 is used for carrying out disease statistics analysis by taking the bridge structure topology, the deployment climate characteristics and the deployment geographic characteristics as constraints, so as to generate the disease exposure degree and the high-frequency disease type of a high-frequency disease area;
A main pattern configuration module 30, wherein the main pattern configuration module 30 is configured to configure an apparent disease identification main pattern of the high-frequency disease region based on the high-frequency disease type in which the disease exposure is greater than or equal to a disease exposure threshold;
A high-frequency disease deleting module 40, wherein the high-frequency disease deleting module 40 is used for deleting the high-frequency disease type from the first disease type set to obtain a second disease type set;
the auxiliary model configuration module 50 is used for constructing an apparent disease identification auxiliary model based on the second disease type set and combining a twin neural network;
the identification area generating module 60 is configured to combine the apparent disease identification main model and the apparent disease identification auxiliary model, construct an apparent disease identification component, process a bridge monitoring image, generate a bridge disease identification area, and send the generated bridge disease identification area to a user side.
Further, the system also comprises a disease exposure generating module for executing the following operation steps:
Taking the bridge structure topology, the deployment climate characteristics and the deployment geographic characteristics as constraints, and collecting bridge disease record data based on a bridge cloud database, wherein any one piece of the bridge disease record data comprises a bridge disease record position and a bridge disease record type;
based on the bridge defect recording type, performing position expansion on the bridge defect recording position in the bridge structure topology to generate a bridge defect recording area;
Clustering the bridge defect recording areas based on an area intersection area ratio threshold value to generate a bridge defect recording area clustering result;
Sorting based on the clustering result of the bridge disease recording area to generate the high-frequency disease area and the high-frequency disease type, wherein the high-frequency disease area and the high-frequency disease type are in one-to-one correspondence;
and carrying out exposure degree analysis on the high-frequency disease area to generate the disease exposure degree, wherein the disease exposure degree is equal to the area high-frequency factor of the high-frequency disease area divided by an area high-frequency factor threshold value.
Further, the system also comprises a bridge defect recording area generating module for executing the following operation steps:
Dividing the bridge disease record type into a corrosive disease type and a non-corrosive disease type;
performing position expansion based on the bridge disease recording position according to the region expansion radius to generate a first bridge disease recording region, wherein the first bridge disease recording region and the corrosion disease type are stored in an associated mode;
Performing position expansion based on the bridge defect recording position according to the region expansion radius and the element splicing boundary to generate a second bridge defect recording region, wherein the second bridge defect recording region and the non-corrosive type are stored in an associated mode;
And adding the second bridge defect recording area and the first bridge defect recording area into the bridge defect recording area.
Further, the system also comprises a clustering result generation module for executing the following operation steps:
Performing intersection analysis on the bridge disease recording areas in pairs to generate area intersection areas and area non-intersection areas which are in one-to-one correspondence;
calculating the sum of the area intersection area and the area non-intersection area to generate an area summation area;
calculating the ratio of the area of the intersection of the areas to the sum area of the areas, and setting the ratio as the area ratio of the intersection of the areas;
And aggregating the bridge defect recording areas with the area intersection area ratio being larger than or equal to the area intersection area ratio threshold value into one type, and generating a clustering result of the bridge defect recording areas.
Further, the system also comprises a high-frequency disease type generating module for executing the following operation steps:
Traversing the clustering result of the bridge disease recording area, and counting area high-frequency factors, wherein the area high-frequency factors are equal to the number of the areas in the class;
Traversing the clustering result of the bridge disease recording area, and counting the high-frequency factors of the diseases, wherein the high-frequency factors of the diseases are equal to the occurrence frequency of each disease type in the class, and the high-frequency factors of the area and the high-frequency factors of the diseases are in one-to-many association;
Performing region fusion on the clustering result of the bridge disease recording region with the region high-frequency factor larger than or equal to the region high-frequency factor threshold value to generate the high-frequency disease region;
setting the disease type of the high-frequency disease area, wherein the disease high-frequency factor of the high-frequency disease area is larger than or equal to a disease high-frequency factor threshold value, as the high-frequency disease type of the high-frequency disease area.
Further, the system also includes a recognition master model generation module to perform the following operational steps:
the bridge foundation information also comprises appearance shape characteristics and appearance color characteristics;
traversing the high-frequency disease type to acquire a first bridge disease image data set based on the appearance shape characteristic and the appearance color characteristic, wherein the first bridge disease image data set comprises a first bridge disease original image data set and a first bridge disease identification image data set with a high-frequency disease type identification;
The first bridge disease identification image data set is provided with a first bridge disease type identification image data set, a second bridge disease type identification image data set and an Nth bridge disease type identification image data set;
The first bridge disease type identification image data set is used as supervision data, the first bridge disease original image data set is used as input data, and a first disease type identification channel is configured based on a convolutional neural network;
The N-th bridge disease type identification channel is configured based on a convolutional neural network by taking the N-th bridge disease type identification image data set as supervision data and the first bridge disease original image data set as input data;
And connecting the first disease type identification channel to the Nth disease type identification channel as parallel nodes, merging input layers, and generating the apparent disease identification main model.
Further, the system also includes an identification auxiliary model generation module to perform the following operation steps:
Based on the appearance shape features and the appearance color features, traversing the second disease type set to acquire a second bridge disease image dataset, wherein the second bridge disease image dataset comprises a second bridge disease original image dataset, a disease texture feature identification dataset and a color feature identification dataset;
The disease texture feature identification data set is used as supervision data, the second bridge disease original image data set is used as input data, and a texture feature extraction channel is configured based on a convolutional neural network;
The color feature identification data set is used as supervision data, the second bridge disease original image data set is used as input data, and a color feature extraction channel is configured based on a convolutional neural network;
the texture feature extraction channel and the color feature extraction channel are used as parallel nodes to be fully connected, so that a feature extraction channel is generated;
And (3) configuring a feature comparison function:
Wherein, P represents a feature comparison output value, p=1 represents success of comparison, p=0 represents failure of comparison, s c represents color similarity, s t represents texture similarity, s c0 represents a color similarity threshold, and s t0 represents a texture similarity threshold;
configuring a color similarity calculation channel and a texture similarity calculation channel;
Copying the feature extraction channels to construct a first feature extraction channel and a second feature extraction channel, wherein model parameters of the first feature extraction channel and the second feature extraction channel are identical to those of the feature extraction channels;
combining the first feature extraction channel and the second feature extraction channel as parallel channels, constructing an input processing layer, combining the color similarity calculation channel and the texture similarity calculation channel as parallel channels, constructing a comparison processing layer, and constructing an output processing layer based on the feature comparison function;
And sequentially connecting the input processing layer, the comparison processing layer and the output processing layer in series to generate the apparent disease identification auxiliary model.
The foregoing detailed description of an apparent recognition method based on bridge exposure will clearly be known to those skilled in the art, and the apparatus disclosed in this embodiment is relatively simple in description, and the relevant points refer to the description of the method section.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (2)

1. An apparent recognition method based on bridge exposure, which is characterized by comprising the following steps:
loading bridge foundation information, wherein the bridge foundation information comprises bridge structure topology, deployment climate characteristics and deployment geographic characteristics;
Performing disease statistical analysis by taking the bridge structure topology, the deployment climate characteristics and the deployment geographic characteristics as constraints, and generating the disease exposure and the high-frequency disease type of a high-frequency disease area;
configuring an apparent disease recognition master model of the high-frequency disease region based on the high-frequency disease type in which the disease exposure is greater than or equal to a disease exposure threshold;
deleting the high-frequency disease type from the first disease type set to obtain a second disease type set;
Based on the second disease type set, combining a twin neural network to build an apparent disease identification auxiliary model;
Combining the apparent disease identification main model and the apparent disease identification auxiliary model, constructing an apparent disease identification assembly, processing a bridge monitoring image, generating a bridge disease identification area and sending the bridge disease identification area to a user side;
the method for generating the disease exposure and the high-frequency disease type of the high-frequency disease area comprises the following steps of:
Taking the bridge structure topology, the deployment climate characteristics and the deployment geographic characteristics as constraints, and collecting bridge disease record data based on a bridge cloud database, wherein any one piece of the bridge disease record data comprises a bridge disease record position and a bridge disease record type;
based on the bridge defect recording type, performing position expansion on the bridge defect recording position in the bridge structure topology to generate a bridge defect recording area;
Clustering the bridge defect recording areas based on an area intersection area ratio threshold value to generate a bridge defect recording area clustering result;
Sorting based on the clustering result of the bridge disease recording area to generate the high-frequency disease area and the high-frequency disease type, wherein the high-frequency disease area and the high-frequency disease type are in one-to-one correspondence;
performing exposure degree analysis on the high-frequency disease area to generate the disease exposure degree, wherein the disease exposure degree is equal to the area high-frequency factor of the high-frequency disease area divided by an area high-frequency factor threshold;
Based on the bridge defect recording type, performing position expansion on the bridge defect recording position in the bridge structure topology to generate a bridge defect recording area, wherein the bridge defect recording area comprises:
Dividing the bridge disease record type into a corrosive disease type and a non-corrosive disease type;
performing position expansion based on the bridge disease recording position according to the region expansion radius to generate a first bridge disease recording region, wherein the first bridge disease recording region and the corrosion disease type are stored in an associated mode;
Performing position expansion based on the bridge defect recording position according to the region expansion radius and the element splicing boundary to generate a second bridge defect recording region, wherein the second bridge defect recording region and the non-corrosive type are stored in an associated mode;
Adding the second bridge defect recording area and the first bridge defect recording area into the bridge defect recording area;
Based on the threshold value of the area ratio of the intersection areas of the areas, clustering the bridge disease recording areas to generate a clustering result of the bridge disease recording areas, comprising:
Performing intersection analysis on the bridge disease recording areas in pairs to generate area intersection areas and area non-intersection areas which are in one-to-one correspondence;
calculating the sum of the area intersection area and the area non-intersection area to generate an area summation area;
calculating the ratio of the area of the intersection of the areas to the sum area of the areas, and setting the ratio as the area ratio of the intersection of the areas;
Aggregating the bridge disease recording areas with the area intersection area ratio being greater than or equal to the area intersection area ratio threshold value into one type, and generating a clustering result of the bridge disease recording areas;
sorting based on the clustering result of the bridge disease recording area to generate the high-frequency disease area and the high-frequency disease type, wherein the sorting comprises the following steps:
Traversing the clustering result of the bridge disease recording area, and counting area high-frequency factors, wherein the area high-frequency factors are equal to the number of the areas in the class;
Traversing the clustering result of the bridge disease recording area, and counting the high-frequency factors of the diseases, wherein the high-frequency factors of the diseases are equal to the occurrence frequency of each disease type in the class, and the high-frequency factors of the area and the high-frequency factors of the diseases are in one-to-many association;
Performing region fusion on the clustering result of the bridge disease recording region with the region high-frequency factor larger than or equal to the region high-frequency factor threshold value to generate the high-frequency disease region;
Setting the disease type of the high-frequency disease area, wherein the disease high-frequency factor of the high-frequency disease area is larger than or equal to a disease high-frequency factor threshold value, as the high-frequency disease type of the high-frequency disease area;
Wherein configuring an apparent disease recognition master model of the high frequency disease region based on the high frequency disease type with the disease exposure greater than or equal to a disease exposure threshold, comprises:
the bridge foundation information also comprises appearance shape characteristics and appearance color characteristics;
traversing the high-frequency disease type to acquire a first bridge disease image data set based on the appearance shape characteristic and the appearance color characteristic, wherein the first bridge disease image data set comprises a first bridge disease original image data set and a first bridge disease identification image data set with a high-frequency disease type identification;
The first bridge disease identification image data set is provided with a first bridge disease type identification image data set, a second bridge disease type identification image data set and an Nth bridge disease type identification image data set;
The first bridge disease type identification image data set is used as supervision data, the first bridge disease original image data set is used as input data, and a first disease type identification channel is configured based on a convolutional neural network;
The N-th bridge disease type identification channel is configured based on a convolutional neural network by taking the N-th bridge disease type identification image data set as supervision data and the first bridge disease original image data set as input data;
connecting the first disease type recognition channel to the Nth disease type recognition channel as parallel nodes, merging input layers, and generating the apparent disease recognition main model;
based on the second disease type set, combining a twin neural network to build an apparent disease identification auxiliary model, wherein the method comprises the following steps of:
traversing the second disease type set to acquire a second bridge disease image data set based on the appearance shape feature and the appearance color feature, wherein the second bridge disease image data set comprises a second bridge disease original image data set, a disease texture feature identification data set and a color feature identification data set;
The disease texture feature identification data set is used as supervision data, the second bridge disease original image data set is used as input data, and a texture feature extraction channel is configured based on a convolutional neural network;
The color feature identification data set is used as supervision data, the second bridge disease original image data set is used as input data, and a color feature extraction channel is configured based on a convolutional neural network;
the texture feature extraction channel and the color feature extraction channel are used as parallel nodes to be fully connected, so that a feature extraction channel is generated;
And (3) configuring a feature comparison function:
，
wherein, Characterization feature comparison output value, p=Characterization and comparison were successful, p=The comparison is characterized by failure in the process of characterization,The color similarity is characterized by the fact that,The similarity of the texture is characterized by the fact that,The color similarity threshold is characterized and,Characterizing a texture similarity threshold;
configuring a color similarity calculation channel and a texture similarity calculation channel;
Copying the feature extraction channels to construct a first feature extraction channel and a second feature extraction channel, wherein model parameters of the first feature extraction channel and the second feature extraction channel are identical to those of the feature extraction channels;
combining the first feature extraction channel and the second feature extraction channel as parallel channels, constructing an input processing layer, combining the color similarity calculation channel and the texture similarity calculation channel as parallel channels, constructing a comparison processing layer, and constructing an output processing layer based on the feature comparison function;
And sequentially connecting the input processing layer, the comparison processing layer and the output processing layer in series to generate the apparent disease identification auxiliary model.
2. An apparent recognition system based on bridge exposure, for implementing the apparent recognition method based on bridge exposure of claim 1, comprising:
the system comprises a basic information loading module, a bridge information processing module and a bridge information processing module, wherein the basic information loading module is used for loading bridge basic information, and the bridge basic information comprises bridge structure topology, deployment climate characteristics and deployment geographic characteristics;
The disease statistics analysis module is used for carrying out disease statistics analysis by taking the bridge structure topology, the deployment climate characteristics and the deployment geographic characteristics as constraints to generate the disease exposure degree and the high-frequency disease type of a high-frequency disease area;
The main model configuration module is used for configuring an apparent disease identification main model of the high-frequency disease area based on the high-frequency disease type with the disease exposure degree larger than or equal to a disease exposure degree threshold value;
the high-frequency disease deleting module is used for deleting the high-frequency disease type from the first disease type set to obtain a second disease type set;
the auxiliary model configuration module is used for constructing an apparent disease identification auxiliary model based on the second disease type set and combining a twin neural network;
the identification region generation module is used for combining the apparent disease identification main model and the apparent disease identification auxiliary model, constructing an apparent disease identification component, processing a bridge monitoring image, generating a bridge disease identification region and sending the bridge disease identification region to a user side.
CN202410030781.2A 2024-01-09 2024-01-09 Apparent recognition method and system based on bridge exposure degree Active CN117851778B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202410030781.2A CN117851778B (en) 2024-01-09 2024-01-09 Apparent recognition method and system based on bridge exposure degree

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202410030781.2A CN117851778B (en) 2024-01-09 2024-01-09 Apparent recognition method and system based on bridge exposure degree

Publications (2)

Publication Number Publication Date
CN117851778A CN117851778A (en) 2024-04-09
CN117851778B true CN117851778B (en) 2024-07-09

Family

ID=90539562

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202410030781.2A Active CN117851778B (en) 2024-01-09 2024-01-09 Apparent recognition method and system based on bridge exposure degree

Country Status (1)

Country Link
CN (1) CN117851778B (en)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110070008A (en) * 2019-04-04 2019-07-30 中设设计集团股份有限公司 Bridge disease identification method adopting unmanned aerial vehicle image
CN115713488A (en) * 2022-10-27 2023-02-24 浙江大学 Bridge apparent disease pixel level identification method and system based on instance segmentation

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115995056A (en) * 2023-03-22 2023-04-21 南京航空航天大学 Automatic bridge disease identification method based on deep learning
CN116665125A (en) * 2023-05-26 2023-08-29 安徽省高速公路试验检测科研中心有限公司 Intelligent recognition system for bridge tunnel diseases

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110070008A (en) * 2019-04-04 2019-07-30 中设设计集团股份有限公司 Bridge disease identification method adopting unmanned aerial vehicle image
CN115713488A (en) * 2022-10-27 2023-02-24 浙江大学 Bridge apparent disease pixel level identification method and system based on instance segmentation

Also Published As

Publication number Publication date
CN117851778A (en) 2024-04-09

Similar Documents

Publication Publication Date Title
CN109118479B (en) Capsule network-based insulator defect identification and positioning device and method
CN108898662B (en) Point cloud data-based pipeline facility BIM (building information modeling) model automatic reconstruction method
CN110703057B (en) Power equipment partial discharge diagnosis method based on data enhancement and neural network
CN111353413A (en) Low-missing-report-rate defect identification method for power transmission equipment
CN108711148B (en) Tire defect intelligent detection method based on deep learning
CN113742327B (en) Automatic rainfall station abnormal value screening method based on rain-measuring radar data
CN117575165B (en) Intelligent patrol management method and system for digital power distribution network
Salari et al. Pavement distress detection and severity analysis
CN117115148B (en) Chip surface defect intelligent identification method based on 5G technology
CN115641327B (en) Building engineering quality supervision and early warning system based on big data
CN114998251B (en) Air multi-vision platform ground anomaly detection method based on federal learning
CN112115770A (en) Method and system for identifying autonomous inspection defects of unmanned aerial vehicle of overhead line
CN117152149B (en) Edge smoothness detection method and system for wafer dicing
CN110794254A (en) Power distribution network fault prediction method and system based on reinforcement learning
CN105844334A (en) Radial basis function neural network-based temperature interpolation algorithm
Zhang et al. Pavement crack detection based on deep learning
CN117851778B (en) Apparent recognition method and system based on bridge exposure degree
CN117557726A (en) Three-dimensional visualization-based power distribution network point cloud model construction system
CN117036429A (en) Infrared-visible light heterologous image registration method for photovoltaic defect detection
CN116543223A (en) Neural network optimized lower tree detection method
CN115830302A (en) Multi-scale feature extraction and fusion power distribution network equipment positioning identification method
CN116342422A (en) Defect identification method based on wafer map denoising
CN114202536A (en) Method for extracting disaster environment features based on AT-EnFCM-SCNN
CN117114197B (en) Symptom migration range identification method based on deep learning
Feng et al. Point Cloud Extraction of Tower and Conductor in Overhead Transmission Line Based on PointCNN Improved

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant