CN114511568A

CN114511568A - Expressway bridge overhauling method based on unmanned aerial vehicle

Info

Publication number: CN114511568A
Application number: CN202210413007.0A
Authority: CN
Inventors: 高腾
Original assignee: Xi'an Bokang Shuoda Network Technology Co ltd
Current assignee: Xi'an Bokang Shuoda Network Technology Co ltd
Priority date: 2022-04-20
Filing date: 2022-04-20
Publication date: 2022-05-17
Anticipated expiration: 2042-04-20
Also published as: CN114511568B

Abstract

The invention relates to the technical field of image processing, in particular to a highway bridge overhauling method based on an unmanned aerial vehicle, which comprises the following steps: acquiring lane lines in each frame of bridge image and grouping the lane lines; acquiring abnormal line segments and corresponding abnormal probabilities, and further acquiring defect probabilities of the bridge images; obtaining distortion rates of each pair of symmetric lines, and classifying all bridge images according to a plurality of distortion rates of each frame of bridge image to obtain a plurality of distortion categories; for each distortion category, grouping all bridge images in the distortion category to obtain a plurality of similar groups; acquiring a sparse matrix of each bridge image, and screening out a difference image according to the similarity between different sparse matrices in each similarity group; and transmitting the bridge image and the difference image with the defect probability larger than the probability threshold value to a server side for defect detection to obtain a defect detection result. The invention can selectively transmit the video frame image, reduce the calculated amount during defect identification and improve the maintenance efficiency.

Description

Expressway bridge overhauling method based on unmanned aerial vehicle

Technical Field

The invention relates to the technical field of image processing, in particular to a highway bridge overhauling method based on an unmanned aerial vehicle.

Background

With the development of the transportation industry in China, the transportation industry is rapidly developed, the driving density and the vehicle load are larger and larger, so that the diseases of the highway bridge are increasingly aggravated, and the cost for maintenance is also gradually increased. The bridge is the throat position in road transport, and therefore road transport has higher requirements on the bridge.

Need unmanned aerial vehicle supplementary when highway bridge overhauls usually, when utilizing unmanned aerial vehicle to overhaul the bridge, what obtain is real-time video image, the data volume is great, carry out bridge deformation through neural network or image processing technique on unmanned aerial vehicle and examine time measuring, the calculated amount is great, the calculation of this kind of magnitude often can not be accomplished to the data processing platform that unmanned aerial vehicle carried on, unmanned aerial vehicle's duration has also been reduced simultaneously, consequently, prior art is when using unmanned aerial vehicle to carry out highway bridge to overhaul, most all data that will obtain all compress the transmission, transmit for the receiving terminal, and because the data volume is great, transmission efficiency often is lower.

Disclosure of Invention

In order to solve the technical problems, the invention aims to provide an unmanned aerial vehicle-based highway bridge overhauling method, which adopts the following technical scheme:

an embodiment of the invention provides a highway bridge overhauling method based on an unmanned aerial vehicle, which comprises the following steps:

carrying out linear detection on bridge images acquired by an unmanned aerial vehicle to obtain lane lines in each frame of bridge image, and grouping the lane lines according to the length of the line segments;

obtaining abnormal line segments according to the number of the line segments of each group, and obtaining corresponding abnormal probability according to the length of the line segments; taking the maximum abnormal probability in each frame of bridge image as the defect probability of the bridge image;

the method comprises the steps of obtaining distortion angles of a plurality of pairs of symmetric lines which are symmetric about a bridge center line in each frame of bridge image, further obtaining distortion rate of each pair of symmetric lines, classifying all bridge images according to the distortion rates of each frame of bridge image, and obtaining a plurality of distortion categories;

for each distortion category, acquiring a dictionary matrix of each frame of bridge image, and grouping all bridge images in the distortion category according to the similarity between different dictionary matrices to obtain a plurality of similarity groups;

acquiring a sparse matrix of each bridge image, and screening out a difference image according to the similarity between different sparse matrices in each similarity group; and transmitting the bridge image and the difference image with the defect probability larger than the probability threshold value to a server side for defect detection to obtain a defect detection result.

Preferably, the method for detecting the lane line comprises the following steps:

and extracting the bridge in the bridge image by removing the background, performing threshold segmentation on the image without the background to obtain a binary image, converting the binary image into a Hough space, and taking the extracted straight line as a lane line.

Preferably, the grouping the lane lines according to the segment lengths includes:

and sorting the detected line segment lengths of the lane lines according to size, and dividing the line segment lengths into a plurality of orders of magnitude through multi-threshold segmentation, wherein the lane lines corresponding to each order of magnitude form a group.

Preferably, the method for acquiring the abnormal line segment includes:

and obtaining standard groups of the known road sections and the number of the standard line sections in each group, and respectively comparing whether the number of the line sections of the group with the maximum line section length and the group with the minimum line section length is consistent with the number of the corresponding standard line sections to obtain abnormal line sections.

Preferably, the method for acquiring the anomaly probability includes:

and acquiring the average length of all the line segments in the group with the longest line segment length, and regarding each abnormal line segment, taking the ratio of the abnormal line segment to the average length as the abnormal probability.

Preferably, the method for obtaining the distortion rate comprises the following steps:

and acquiring a difference value of the abscissa of the corresponding point of each pair of symmetrical lines in the Hough space as the distortion angle, and taking the ratio of the distortion angle to a preset angle as the distortion rate of the symmetrical lines.

Preferably, the method for acquiring the plurality of distortion categories is as follows:

and obtaining distortion sequences consisting of distortion rates of all symmetric lines in each frame of bridge image, calculating cosine similarity of every two distortion sequences, classifying all bridge images by utilizing the cosine similarity, and obtaining a plurality of distortion categories.

Preferably, the method for acquiring the plurality of similarity groups comprises:

converting the dictionary matrix into a row vector form according to a row vector end-to-end mode, recording the form as a dictionary row vector, calculating the mean value of cosine similarity between each dictionary row vector and other dictionary row vectors as the average similarity of the dictionary row vectors, and performing multi-threshold segmentation on all the average similarities to obtain a plurality of similarity groups.

Preferably, the obtaining of the plurality of similarity groups further comprises the following steps:

and acquiring the maximum row number and the maximum column number of an image matrix corresponding to all the frame bridge images, and expanding the row number of all the frame images into the maximum row number and the column number into the maximum column number by a nearest neighbor interpolation method so as to enable all the images to be the same in size.

Preferably, the difference image obtaining method includes:

for each similarity group, acquiring the sum of the similarity of each bridge image and all other bridge images in the same group as the first similarity of the bridge images, sequencing according to the size of the first similarity, and selecting the bridge images with the preset number as reserved images; and calculating second similarity of the sparse coding matrix of each reserved image and other reserved images in the same group, and taking the bridge image corresponding to the second similarity smaller than the similarity threshold value as the difference image.

The embodiment of the invention at least has the following beneficial effects:

judging the defect probability at the position of the lane line through the abnormity of the lane line; selecting a difference image in each similar group by performing similar grouping on the images with the same distortion; and (4) taking the bridge image with the larger defect probability and the relatively representative difference image as images with larger probability of defects, and transmitting the images to the server side for defect detection. According to the invention, the images with larger possible defects are screened out for selective transmission, so that the calculated amount in the data processing process is reduced, and the overhauling efficiency is improved.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions and advantages of the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.

Fig. 1 is a flowchart illustrating steps of a method for repairing a highway bridge based on an unmanned aerial vehicle according to an embodiment of the present invention.

Detailed Description

To further illustrate the technical means and effects of the present invention adopted to achieve the predetermined invention, the following detailed description will be given to the specific implementation, structure, features and effects of the method for repairing a highway bridge based on an unmanned aerial vehicle according to the present invention with reference to the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" refers to not necessarily the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

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 invention belongs.

The following describes a specific scheme of the unmanned aerial vehicle-based highway bridge maintenance method in detail with reference to the accompanying drawings.

Referring to fig. 1, a flow chart of steps of a method for repairing a highway bridge based on an unmanned aerial vehicle according to an embodiment of the present invention is shown, which includes the following steps:

and S001, carrying out linear detection on the bridge images acquired by the unmanned aerial vehicle to obtain lane lines in each frame of bridge images, and grouping the lane lines according to the length of the line segments.

The method comprises the following specific steps:

1. and acquiring a lane line in the bridge image.

The method comprises the steps of extracting a bridge in a bridge image by removing a background, carrying out threshold segmentation on the image after the background is removed to obtain a binary image, converting the binary image into a Hough space, and taking an extracted straight line as a lane line.

When the unmanned aerial vehicle overhauls the bridge, a fixed routing inspection path is provided, the background is removed through the DNN semantic segmentation network in the embodiment of the invention, the DNN semantic segmentation network has a simple identification purpose, only the DNN semantic segmentation network is used for distinguishing the foreground and the background, the network parameters are few, and the foreground refers to the bridge. The DNN network is located in an unmanned aerial vehicle embedded system, is trained in advance, can be directly used for semantically segmenting images, obtains bridge foreground images, and is small in calculation amount, high in running speed and free of calculation pressure on the unmanned aerial vehicle.

The training process of the DNN semantic segmentation network comprises the following steps:

overlooking the bridge image data set of the collection as training set, the style of the bridge is various, the pixel that needs to be cut apart is divided into two types altogether, namely the training set corresponds the label marking process and is: and in the semantic label of the single channel, the label of the pixel at the corresponding position belonging to the background class is 0, and the label of the pixel belonging to the bridge is 1. The task of the network is to classify and the loss function used is a cross-entropy loss function.

The mask image obtained by semantic segmentation is multiplied by the original image, and the obtained image only contains the image of the bridge, so that the interference of the background is removed.

The lane lines on the expressway are straight lines, when the defects on the surface of the bridge are at the positions of the lane lines, the lane lines can deform, and the deformation degree of the lane lines can reflect the defect conditions of the bridge at the positions of the lane lines. By detecting these pieces of straight line information, the degree of bridge deformation is determined from the overall situation.

Firstly, performing threshold segmentation on a gray level image through the Otsu method to obtain a binary image, then performing Hough change on the binary image to obtain a Hough parameter space, wherein a plurality of highlight points exist in the Hough parameter space, and each highlight point corresponds to a straight line, namely a lane line.

2. And grouping the lane lines according to the length of the line segment.

The lane lines on the highway bridge are parallel solid lines and dotted lines, the number of highlight points in the Hough space represents the number of straight lines, and the voting values of two highlight points represent the lengths of the straight lines, so that the voting value of the highlight points corresponding to the dotted lines in the Hough parameter space is smaller than the voting value of the highlight points corresponding to the solid lines in the Hough parameter space.

The line segments with the same length in the image space correspond to two highlight points with the same voting value in the Hough parameter space, so that the probability of whether the straight line in the image space is deformed can be obtained by calculating the similarity of the voting values of the highlight points in the Hough parameter space. Since there are solid and dashed lines in parallel lines, there are two kinds of highlight point values in a normal case without defects.

The voting values of the points in the Hough parameter space are arranged in a descending order, different orders of magnitude are obtained through multi-threshold segmentation, and the mean value of the number of elements corresponding to each order of magnitude is used as the value of each element.

For example, one order of magnitude obtained by multi-threshold segmentation is (40, 41, 41, 42, 43, 43, 44), the average value 42 of 7 elements in the order of magnitude is taken as the value of each element, the final order of magnitude is (42, 42, 42, 42, 42, 42, 42, 42), and the lane line corresponding to each order of magnitude is a group.

The purpose of multi-threshold segmentation is to make the elements with similar numbers be of the same order of magnitude, thereby obtaining different orders of magnitude, which are called as different classes. And counting the number in different categories to obtain an abnormal straight line.

The specific method of multi-threshold segmentation is to perform multi-threshold segmentation on the sequence according to the Fisher criterion by using the principle that the inter-class variance is maximum and the intra-class variance is minimum.

S002, obtaining abnormal line segments according to the number of the line segments of each group, and obtaining corresponding abnormal probability according to the length of the line segments; and taking the maximum abnormal probability in each frame of bridge image as the defect probability of the bridge image.

The method comprises the following specific steps:

1. and acquiring abnormal line segments according to the number of the line segments in each group.

When the unmanned aerial vehicle detects the road section, the route is planned in advance, so that the name of the road section overhauled at each moment and the condition of the corresponding road section can be obtained, namely the number of the solid lines and the broken lines on the bridge at each moment can be obtained in advance, the number of the solid lines is represented by M, and the number of the broken lines is represented by N. The standard groups obtained under normal conditions are divided into two groups, one group is a solid line, the other group is a dotted line, the number of the standard line segments in the solid line group is M, and the number of the standard line segments in the dotted line group is N.

The line segments with M + N before ranking are taken from the sequence with the line segment length arranged from big to small, and are grouped by multi-threshold segmentation, and the ideal classes are as follows: the M before ranking is used as a group corresponding to the real line group; as one group, ranked at (M +1, N), corresponding to the group of virtual lines. In practice, due to defects, the number of groups is not two, and the number of segments in the groups is not consistent, which indicates that there may be abnormal segments.

Comparing the number M of the groups with the maximum length of the line segment, namely the first group, with the number M, and if M is smaller than M, indicating that the elements in the first group are divided into the second group, wherein the divided elements may be the class change caused by defects; then comparing the number N of the group with the minimum length of the line segment, namely the last group with N, if N is less than N, indicating that the positions of the solid line and the dotted line have a high probability of having defects, and needing to find abnormal line segments.

If the number M of line segments within the first group is less than M, all line segments ranked after (M-M) are anomalous line segments; meanwhile, if the number N of the line segments in the last group is less than N, all the line segments ranked at (M +1, M + N) are abnormal line segments; if the number of segments N in the last group is greater than N, then all segments ranked at (M +1, M + N-N) are anomalous segments.

2. And acquiring the defect probability of the bridge image.

And acquiring the average length of all the line segments in the group with the longest line segment length, taking the ratio of the average length to each abnormal line segment as the abnormal probability, and taking the maximum abnormal probability in each frame of bridge image as the defect probability of the bridge image.

And S003, acquiring distortion angles of each pair of symmetric lines which are symmetric about the center line of the bridge in each frame of bridge image, further acquiring distortion rates of each pair of symmetric lines, and classifying all bridge images according to a plurality of distortion rates of each frame of bridge image to obtain a plurality of distortion categories.

When the unmanned aerial vehicle carries out aerial photography on the bridge, obstacles need to be avoided, so the heights when different positions of the bridge are shot are possibly different, the near-far distortion degree of the bridge in different video frame images is different, all bridge images are classified according to a plurality of distortion rates of each frame of bridge image, and a plurality of distortion categories are obtained.

The method comprises the following specific steps:

1. the distortion rate of each pair of symmetry lines is obtained.

And acquiring the difference value of the abscissa of the corresponding point of each pair of symmetrical lines in the Hough space as a distortion angle, and taking the ratio of the distortion angle to a preset angle as the distortion rate of the symmetrical lines.

The midline of the bridge is taken as a dividing line, and the parallel lines are symmetrical about the midline. The distance value of the abscissa of two highlight points corresponding to the symmetric line in the Hough space represents the degree of near-far distortion of the symmetric line in the image, and the larger the distance value is, the larger the distortion is. Obtaining each pair of symmetrical lines in Hough spaceThe difference of the abscissa of the corresponding point is taken as the distortion angle

A distortion rate q with a ratio of the distortion angle to a preset angle as a symmetry line:

as an example, the preset angle is 90 degrees in the embodiment of the present invention. The larger the distortion rate, the larger the distortion of the image caused by the near and far.

2. And classifying all the bridge images according to the distortion rates of each frame of bridge image to obtain a plurality of distortion categories.

The distortion rates of all symmetric lines in each frame of bridge image form a distortion sequence of the bridge image, cosine similarity between the distortion sequences corresponding to each two frames of bridge images is calculated, firstly, multi-threshold segmentation is carried out on all cosine similarity to obtain a plurality of categories, the cosine similarity in each category is similar, for each category, distortion sequences which do not calculate the cosine similarity mutually in the corresponding distortion sequences are selected, and the category is corrected according to the cosine similarity between the selected distortion sequences to obtain a final classification result.

For example, five bridge images of ABCDE are in the same category,

、

、

、

the similarity is high, the ABCDE is divided into a class, wherein the similarity between the ABCE is calculated and is high, the similarity between the C and the D is not calculated with other three bridge images, and further taking A as an example, the similarity is calculated

、

If both the obtained similarities are low, the similarity between the CD and the ABE is low, but the similarities between the C and the D are high, the ABE is divided into one class, and the C and the D are independently divided into one class, so that the correction of the ABCDE is completed.

And step S004, acquiring a dictionary matrix of each frame of bridge image for each distortion category, and grouping all bridge images in the distortion category according to the similarity between different dictionary matrices to obtain a plurality of similarity groups.

The method comprises the following specific steps:

1. for each distortion category, a dictionary matrix is obtained in which each frame of bridge image is acquired.

In the embodiment of the invention, the dictionary matrix is obtained through K-SVD. K-SVD is a dictionary representation method, which decomposes the original image matrix into the product of a dictionary matrix and a sparse coding matrix.

The dictionary matrix represents the characteristic information of the images, namely the most essential information, and the difference of the dictionary matrices of the two images is equivalent to the difference of the articles in the two images; the sparse matrix represents a combination of different features in the image. The dictionary matrix determines the essential information of the image, namely, which information is contained, and the sparse matrix represents the combination mode of the information.

2. And acquiring the maximum row number and the maximum column number of an image matrix corresponding to all the frame bridge images, and expanding the row number of all the frame images into the maximum row number and the column number into the maximum column number by a nearest neighbor interpolation method so as to enable all the images to be the same in size.

The similarity between dictionary matrixes is calculated by firstly requiring the consistency of the sizes of the dictionary matrixes, wherein the sizes of the dictionary matrixes mainly depend on two parameters, one parameter is the line number of an image matrix, the other parameter is a parameter set manually, the parameter set manually is a set fixed value, and the consistency of the parameter is kept. After semantic segmentation, the number of rows of the image matrix is not necessarily the same, so that the number of rows and columns of the largest different video frame image needs to be calculated, and then the sizes of different image matrices are kept consistent by a nearest neighbor interpolation method.

And calculating to obtain the maximum line number and the maximum column number of all video frame image matrixes, and expanding the line number and the column number of all video frame images into the corresponding maximum line number and maximum column number by using a nearest neighbor interpolation method to obtain bridge images of different video frames with the same size.

3. A plurality of similarity groups in each distortion category is obtained.

The dictionary matrixes with similar similarity are classified into one class by a multi-threshold segmentation method, namely, the bridge images with the similar dictionary matrixes are classified into one class, and different similarity groups are obtained.

S005, acquiring a sparse matrix of each bridge image, and screening out a difference image according to the similarity between different sparse matrices in each similarity group; and transmitting the bridge image and the difference image with the defect probability larger than the probability threshold value to a server side for defect detection to obtain a defect detection result.

The method comprises the following specific steps:

1. and screening out the difference images.

For each similarity group, acquiring the sum of the similarity of each bridge image and all other bridge images in the same group as the first similarity of the bridge images, sequencing according to the size of the first similarity, and selecting the bridge images with the preset number as reserved images; and calculating second similarity of the sparse coding matrix of each reserved image and other reserved images in the same group, and taking the bridge image corresponding to the second similarity smaller than the similarity threshold value as a difference image.

As the number of video frame images is too large, a preset number of reserved images are selected from each similarity group to represent the similarity group, and as an example, the method for calculating the similarity in the embodiment of the present invention is cosine similarity, and the preset number is 100.

The image with the smaller similarity represents that the image has larger difference with other images, is more unique and is more likely to have defects; the image with the second similarity is characterized in that the image with the second similarity is similar to other images, can be represented by one of the images, and is more likely to be a common image without defects; therefore, the bridge image corresponding to the second similarity smaller than the similarity threshold is selected as the difference image.

In the embodiment of the invention, the similarity threshold is obtained by a maximum inter-class variance method.

2. And screening the image for defect detection.

The bridge image with the defect probability larger than the probability threshold and the difference image are transmitted to a server side for defect detection, the bridge image with the defect probability larger than the probability threshold is likely to be a bridge image with defects at the position of a lane line, and the difference image is an image capable of representing representativeness and is more likely to be a bridge image with defects at other positions.

As an example, the value of the probability threshold in the embodiment of the present invention is 0.6.

After the bridge image is sent to the server, the defects of the bridge image are more carefully identified through modes such as threshold segmentation, edge detection or neural network, the image transmitted to the server has defects with high probability, the occurrence of events that the defects are not detected after the detection is finished is reduced, and the calculated amount is reduced.

In the embodiment of the invention, the neural network is adopted to detect the defects of the transmission image, wherein the defects comprise corner drop and edge loss of the bridge structure caused by vehicle impact and the like; the structural connection of the bridge is damaged by cracking, breaking and the like; the defects of the deep pits on the surface of the bridge and the like are used as labels to train the neural network, and the trained neural network is used for identifying the different defects, so that a large amount of calculation is needed, the defects are difficult to identify on an unmanned aerial vehicle, and the defects need to be transmitted to a server side for identifying and calculating specific defect types.

In summary, in the embodiment of the present invention, the lane lines in each frame of bridge image are obtained by performing straight line detection on the bridge image acquired by the unmanned aerial vehicle, and the lane lines are grouped according to the length of the line segment; obtaining abnormal line segments according to the number of the line segments of each group, and obtaining corresponding abnormal probability according to the length of the line segments; taking the maximum abnormal probability in each frame of bridge image as the defect probability of the bridge image; the method comprises the steps of obtaining distortion angles of a plurality of pairs of symmetric lines which are symmetric about a bridge center line in each frame of bridge image, further obtaining distortion rate of each pair of symmetric lines, classifying all bridge images according to the distortion rates of each frame of bridge image, and obtaining a plurality of distortion categories; for each distortion category, acquiring a dictionary matrix of each frame of bridge image, and grouping all bridge images in the distortion category according to the similarity between different dictionary matrices to obtain a plurality of similarity groups; acquiring a sparse matrix of each bridge image, and screening out a difference image according to the similarity between different sparse matrices in each similarity group; and transmitting the bridge image and the difference image with the defect probability larger than the probability threshold value to a server side for defect detection to obtain a defect detection result. The embodiment of the invention can selectively transmit the video frame image, reduce the calculated amount during defect identification and improve the maintenance efficiency.

It should be noted that: the precedence order of the above embodiments of the present invention is only for description, and does not represent the merits of the embodiments. And specific embodiments thereof have been described above. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.

The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments.

The above-mentioned embodiments are only used for illustrating the technical solutions of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; such modifications and substitutions do not substantially depart from the spirit and scope of the embodiments of the present application and are intended to be included within the scope of the present application.

Claims

1. An unmanned aerial vehicle-based highway bridge overhauling method is characterized by comprising the following steps:

the method comprises the steps of obtaining distortion angles of each pair of symmetric lines which are symmetric about a bridge center line in each frame of bridge image, further obtaining distortion rates of each pair of symmetric lines, classifying all bridge images according to a plurality of distortion rates of each frame of bridge image, and obtaining a plurality of distortion categories;

2. The unmanned aerial vehicle-based highway bridge overhauling method according to claim 1, wherein the detection method of the lane line is as follows:

3. The unmanned-aerial-vehicle-based highway bridge overhauling method of claim 1, wherein grouping the lane lines according to line length comprises:

4. The unmanned aerial vehicle-based highway bridge overhauling method according to claim 3, wherein the abnormal line segment obtaining method comprises the following steps:

5. The unmanned aerial vehicle-based highway bridge overhauling method according to claim 4, wherein the obtaining method of the abnormal probability comprises the following steps:

6. The unmanned aerial vehicle-based highway bridge overhauling method according to claim 1, wherein the distortion rate obtaining method comprises the following steps:

7. The unmanned aerial vehicle-based highway bridge overhauling method according to claim 1, wherein the plurality of distortion categories are obtained by the following steps:

8. The unmanned aerial vehicle-based highway bridge overhauling method according to claim 1, wherein the plurality of similarity groups are obtained by:

9. The unmanned-aerial-vehicle-based highway bridge overhauling method of claim 8 wherein said obtaining a plurality of similarity groups further comprises the steps of:

10. The unmanned aerial vehicle-based highway bridge overhauling method according to claim 1, wherein the difference image obtaining method comprises the following steps: