Road damage intelligent identification method and system based on image analysis

Abstract

The invention relates to the technical field of road damage detection, in particular to an intelligent road damage recognition method and system based on image analysis, and the method comprises the following steps: obtaining road images, screening out substandard images, extracting local features of the images, executing enhancement processing, converting structural parameters, and matching a type template; and combining the vehicle axle load and the damage index to calculate a structure expansion trend, and generating damage expansion risk information. According to the method, the accuracy of damage type identification is improved by combining geometric parameters such as edge length, curvature range and closed region area value and performing scale conversion based on pixel resolution, and an evaluation mechanism of the influence of vehicle load on the damaged region is established by combining traffic flow and vehicle axle load data. The expansion trend is judged through the traffic frequency and the impact weight, the damage development situation is effectively predicted, the accident frequency, the speed limit information and the traffic obstruction factor are introduced into priority ranking calculation, and the urgency recognition capability of road maintenance is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of road damage detection, and particularly to an intelligent recognition method and system for highway damage based on image analysis. Background Art

[0002] The technical field of road damage detection mainly involves the status monitoring and evaluation of infrastructure such as highways and urban roads. The purpose is to detect and identify damage on the road surface, such as cracks, potholes, wear, etc., through efficient and accurate methods. The technologies in this field usually rely on means such as image processing, computer vision, and artificial intelligence algorithms to detect, classify, and evaluate road damage through an automated system. With the continuous progress of technology, traditional manual detection methods have gradually been replaced by intelligent detection systems based on image analysis and machine learning, improving the efficiency, accuracy, and operability of detection. The application of such technologies not only improves the efficiency of road maintenance work but also provides a data basis for preventive maintenance and decision support.

[0003] Among them, the intelligent recognition method for highway damage aims to automatically identify and evaluate highway road damage through detection technology. By using high-resolution imaging devices to collect image data of the highway surface and combining deep learning or computer vision algorithms for processing, it can accurately identify the damaged areas on the road and their types, such as cracks, potholes, peeling, etc. Through image processing and analysis, the system can provide detailed information on road damage, thereby providing timely and effective decision support for relevant management departments. The wide application of this method helps to improve the intelligent level of road maintenance and achieve precise and efficient road repair and management.

[0004] When the image quality is not high, traditional detection methods cannot accurately extract the details of damage, resulting in inaccurate evaluation of the damage type and severity, affecting the overall judgment effect. In the process of extracting the characteristics of damaged areas, it mainly relies on fixed templates or classification networks, lacking boundary enhancement and contrast optimization for the characteristics of different regions in the image, resulting in blurred edges of complex structures, making it difficult to accurately extract geometric forms such as closed areas and curvatures, and not quantitatively evaluating the development trend of damage, making it difficult to support the formulation of long-term maintenance strategies. At the level of damage treatment decision-making, current means generally rely on single-index judgment, failing to comprehensively consider multi-dimensional traffic data, making it difficult to accurately identify high-risk sections that need to be processed first, resulting in insufficient judgment dimensions and being delayed in processing, increasing the risk of accidents and reducing the traffic safety of the road. Summary of the Invention

[0005] The purpose of the present invention is to solve the deficiencies existing in the prior art and propose an intelligent recognition method and system for highway damage based on image analysis.

[0006] To achieve the above object, the present invention adopts the following technical solution: An intelligent road damage recognition method based on image analysis, comprising the following steps:

[0007] S1: Obtain the collected road surface images, calculate the image clarity evaluation index, screen out the images that do not meet the quality reference value conditions, record the image path and identification number, and generate the image clarity screening result;

[0008] S2: Based on the image clarity screening result, obtain the local contrast value and spatial frequency distribution value in each regional image, perform edge sharpening operation on the image according to the boundary strength value and contrast enhancement weight, and identify the damaged edge curve, notch shape and closed contour feature in the enhanced image to generate the damaged area structure feature information;

[0009] S3: According to the damaged area structure feature information, call the pixel resolution coefficient of the road image, convert the area and boundary length of each damaged area in the actual road surface, and perform matching judgment based on the damaged structure form and the known damaged type boundary form to identify the damaged type and obtain the damaged type classification information;

[0010] S4: Based on the damaged type classification information, extract the average axle load value, traffic frequency and the corresponding frequency value of the highest axle load in each time period, and perform an expansion trend evaluation according to the cumulative action weight value of the vehicle load impact on each damaged structure to generate the damaged expansion risk information.

[0011] The present invention is improved in that the image clarity screening result includes the image path index number, image clarity evaluation value, quality determination status and image validity mark, the damaged area structure feature information is specifically the edge contour coordinate group, area closed state, pixel gradient distribution and structure boundary strength, the damaged type classification information includes the actual crack area, structure boundary length, damaged form number, type attribution label, and the damaged expansion risk information specifically refers to the load impact parameter, traffic frequency coefficient, structure expansion trend value and risk level factor.

[0012] The present invention is improved in that the steps for obtaining the image clarity screening result are specifically as follows:

[0013] S111: Obtain the collected road surface images, extract the pixel matrices in the rut area, joint area and edge area of the image, call the image pixel values to construct a grayscale value distribution histogram, and synchronously extract the noise distribution density value at the gray level mutation position in the image signal to obtain the basic pixel characteristic data;

[0014] S112: Based on the basic pixel characteristic data, call the gray level gradient change degree and edge response value of the image boundary area to construct a gradient intensity equilibrium vector group and an edge response value group, and use the formula:

[0015]

[0016] Obtain the clarity evaluation index of the image through calculation, and perform position judgment on the clarity evaluation index and the set image quality reference value interval, screen out the image samples that exceed the lower limit of the reference value, and obtain the image validity determination information;

[0017] Among them, S i represents the clarity evaluation index of the i-th image, and G i,j represents the gray gradient intensity value of the j-th region in the i-th image, and E i,j represents the edge response intensity value of the j-th region in the i-th image, and N i,j represents the noise density value in the j-th region of the i-th image, and n represents the total number of regions into which the image is divided;

[0018] S113: According to the image validity determination information, extract the image paths and identification numbers that meet the conditions of the image quality reference interval, set quality level labels for the image set, and establish the image clarity screening result.

[0019] The improvement of the present invention is that the step of obtaining the damaged area structure feature information is specifically as follows:

[0020] S211: Based on the image clarity screening result, extract the pixel matrices of the rut edge region, the central partition region, and the staggered joint region in the screened image, locate the horizontal and vertical pixel distribution position indexes in each region, calculate the adjacent pixel gray difference sequence and the arrangement angle gradient sequence, and statistically calculate the gray mean difference and the direction distribution stability of each region to obtain the region gray and direction statistical values;

[0021] S212: According to the region gray and direction statistical values, extract the gray mutation gradient value, the contrast enhancement amplitude value, and the boundary line angle offset value of each structure boundary line in the image, perform boundary response reconstruction operations on the edges of each line segment, and at the same time perform local brightness increment operations on the pixel groups with brightness lower than the median value of the overall image brightness. Use the formula:

[0022]

[0023] Perform calculations to obtain the structure offset response coefficient;

[0024] Among them, R s represents the structure offset response coefficient, and D k represents the normalized value of the gray mutation gradient of the k-th boundary line, and C k represents the normalized amplitude value of the contrast enhancement of the k-th segment, and θ k represents the normalized value of the angle offset of the k-th boundary line, and M krepresents the normalized value of the brightness increment of the k-th segment boundary line, α is the contrast adjustment coefficient, β is the offset suppression factor, and N R is the number of structural boundary line segments in the image;

[0025] S213: Invoke the structural offset response coefficient, extract the pixel line segment group in the boundary structure segment where the closed trend value is greater than the offset response mean value, calculate the closed arc distance between the endpoints and the segment intersection frequency, determine whether a continuous closed path is formed based on the curve density distribution, label the regional contour structure forming the closed boundary, and establish the damaged area structure feature information.

[0026] The improvement of the present invention is that the obtaining step of the damage type classification information is specifically as follows:

[0027] S311: Based on the damaged area structure feature information, extract the edge pixel path corresponding to each area, invoke the edge pixel index value sequence, calculate the cumulative value of the Euclidean distance between adjacent pixel points, count the total edge path length, and invoke the pixel unit number within the corresponding closed contour and the pixel resolution conversion coefficient to calculate the actual area of the closed area, and obtain the damaged area and the boundary quantization value;

[0028] S312: According to the damaged area and the boundary quantization value, extract the total edge length value, the closed area value, and the contour average curvature value of each damaged area, invoke the pre-set damaged form feature matching template, and use the formula:

[0029]

[0030] Perform operations to obtain the structural form matching difference value of the damaged area to be classified;

[0031] where S z represents the structural form matching difference value, A d is the normalized area value of the area to be recognized, A m is the standard area value of the matching template, L d is the boundary length value of the area to be recognized, L m is the boundary length value of the matching template, C d is the average curvature value of the area to be recognized, C m is the reference curvature value of the matching template, and γ is the curvature difference adjustment coefficient;

[0032] S313: Invoke the structural form matching difference value of the damaged area to be classified, extract the template label number corresponding to the minimum structural form matching difference value, identify the corresponding damage type, screen the number set with a difference value less than the structural form tolerance threshold in the damaged area, establish the type index mapping relationship and generate the annotation value, and establish the damage type classification information.

[0033] The improvements of the present invention are as follows. The step of obtaining the damage expansion risk information is specifically as follows:

[0034] S411: Based on the damage type classification information, extract the highway section numbers associated with each damaged area, obtain the traffic flow data and vehicle axle load data within a continuous time period obtained by the vehicle detection equipment corresponding to the section, count the vehicle passing frequencies within each time period, calculate the average axle load value and the maximum axle load occurrence frequency within each time period, and establish a traffic load statistical value;

[0035] S412: According to the traffic load statistical value, call the area quantization value and depth estimation value of the corresponding damaged area, extract the damage type and structure code, match the preset load impact structure action coefficient, perform a structural cumulative impact assessment operation, and use the formula:

[0036]

[0037] Calculate the damage impact response value corresponding to each time period under each type of structure, and obtain the structural impact response coefficient;

[0038] Among them, R d represents the structural impact response coefficient, H a,t is the average axle load value in the t-th time period, H m,t is the maximum axle load value in the t-th time period, S z,t is the vehicle passing frequency in the t-th time period, A d,t is the area quantization value of the area in the t-th time period, D d,t is the damage depth normalization value in the t-th time period, λ is the depth enhancement factor, and T is the total number of time periods;

[0039] S413: Call the structural impact response coefficient, match the structure code, response value and threshold level table corresponding to each damage type for level matching, assign an expansion risk marker label and output a number comparison table, and establish damage expansion risk information.

[0040] The improvements of the present invention are as follows. The method further includes;

[0041] S5: Call the damage expansion risk information, combine the lane traffic flow ratio, speed limit section and accident record section number, evaluate the processing priority of road damage, perform a processing sequence sorting, and obtain road damage processing sorting information;

[0042] The road damage processing sorting information includes a priority score value, a road section identification number, an obstacle weight parameter, and a sorting output index number.

[0043] The improvements of the present invention are as follows. The step of obtaining the road damage processing sorting information is specifically as follows:

[0044] S511: Invoke the damaged expansion risk information, extract the corresponding road section number, obtain the lane traffic flow proportion, speed limit section identifier, and historical accident record information corresponding to the road section, match the impact degree level value in the accident record, combine it with the speed limit section number, calculate the accident interference level value, convert it into a standardized index, and adjust the penalty factor in combination with the traffic flow proportion to obtain the traffic interference coefficient value;

[0045] S512: According to the traffic interference coefficient value, extract the corresponding damaged area structure type code and area normalization value, invoke the structure type risk factor weight, and perform a comprehensive priority score calculation according to the predefined risk assessment expression, using the formula:

[0046]

[0047] Perform operations to obtain the weighted priority score value of each damaged area;

[0048] Among them, P u represents the weighted priority score value, R u is the structural risk factor of the area, A u is the area damage area normalization value, F u is the traffic interference coefficient value of the area, B u is the accident impact level index of the area, and η is the accident penalty enhancement factor;

[0049] S513: Based on the weighted priority score value of each damaged area, perform a sorting process on the score values of each damaged area, combine the road section number and the corresponding sorting result to construct a processing queue index table, and establish road damage processing sorting information.

[0050] An intelligent highway damage identification system based on image analysis, the intelligent highway damage identification system based on image analysis is used to implement the intelligent highway damage identification method based on image analysis, and the system includes:

[0051] The image screening and processing module acquires the collected road surface image, calculates the image clarity evaluation index, screens out the images that do not meet the quality reference value conditions, and records the image path and identification number to generate the image clarity screening result;

[0052] The damage feature recognition module, based on the image clarity screening result, obtains the local contrast value and spatial frequency distribution value in each area image, performs edge sharpening operations on the image according to the boundary strength value and contrast enhancement weight, and identifies the damage edge curve, notch shape, and closed contour features in the enhanced image to generate the damaged area structure feature information;

[0053] The damage type evaluation module calculates the area and boundary length of each damaged area in the actual road surface according to the structural feature information of the damaged area, and makes a matching judgment based on the structural form of the damage and the known boundary form of the damage type to identify the damage type and obtain the damage type classification information;

[0054] Based on the damage type classification information, the damage expansion analysis module evaluates the expansion trend according to the cumulative action weight value of vehicle load impact on each damaged structure and generates the damage expansion risk information;

[0055] The damage treatment evaluation module calls the damage expansion risk information, combines the lane traffic flow ratio, speed limit section and accident record section number, evaluates the treatment priority of road damage, sorts the treatment order, and obtains the road damage treatment sorting information.

[0056] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0057] In the present invention, by quantitatively evaluating and screening the clarity of road surface images, low-quality images are excluded, identification errors caused by blurred or noisy images are avoided, the accuracy of subsequent image analysis is enhanced, pixel distribution analysis and edge sharpening processing are performed on typical areas in the screened images, and by calculating local contrast and spatial frequency, the clarity of boundary features is enhanced, and the extraction accuracy of structural features of damaged areas is improved. In the process of damaged area identification, geometric parameters such as edge length, curvature range, and closed area value are combined, and scale conversion is performed based on pixel resolution, so as to accurately quantify the damaged area and boundary length in the actual road, improve the accuracy of damage type identification, combine traffic flow and vehicle axle load data, establish an evaluation mechanism for the impact of vehicle load on damaged areas, judge the expansion trend through traffic frequency and impact weight, effectively predict the development trend of damage, introduce accident frequency, speed limit information and traffic obstruction factors into the priority sorting calculation, and perform weighting in combination with damage structure risk factors and area indicators, so as to scientifically judge the order of treatment and enhance the ability to identify the urgency of road maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 is the method flow chart of the present invention;

[0059] Figure 2 is the flow chart of obtaining the image clarity screening result of the present invention;

[0060] Figure 3 is the flow chart of obtaining the structural feature information of the damaged area of the present invention;

[0061] Figure 4 is the flow chart of obtaining the damage type classification information of the present invention;

[0062] Figure 5 Flow chart for obtaining damage expansion risk information in the present invention;

[0063] Figure 6 Flow chart for obtaining road damage treatment sorting information in the present invention. Specific embodiments

[0064] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0065] In the description of the present invention, it should be understood that the orientation or positional relationships indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. are based on the orientation or positional relationships shown in the accompanying drawings. These are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of the present invention. In addition, in the description of the present invention, the meaning of "a plurality of" is two or more, unless otherwise specifically defined.

[0066] Please refer to Figure 1 , the present invention provides a technical solution: an intelligent road damage identification method based on image analysis, including the following steps:

[0067] S1: Obtain the collected road surface image, call the gray value distribution histogram of the damaged area in the image and the overall noise distribution density value of the image, compare the degree of gray gradient change and the boundary pixel edge response value, calculate the image clarity evaluation index, compare the set image quality reference value, screen out the images that do not meet the quality reference value conditions, and record the image path and identification number to generate the image clarity screening result;

[0068] S2: Based on the image clarity screening result, for the screened images, perform boundary positioning processing on the pixel point distribution in the rut edge area, the central partition area and the staggered joint area in the image, obtain the local contrast value and the spatial frequency distribution value in each area image, perform edge sharpening operation on the image according to the boundary strength value and the contrast enhancement weight, and perform local brightness adjustment processing on the boundary area, and identify the damaged edge curve, notch shape and closed contour feature in the enhanced image to generate the damaged area structure feature information;

[0069] S3: According to the structural feature information of the damaged area, based on the edge length, curvature range and enclosed area value of the damaged area, call the pixel resolution coefficient of the road image, convert the area and boundary length of each damaged area in the actual road surface, and make a matching judgment according to the structural form of the damage and the known boundary form of the damage type to identify the damage type and obtain the damage type classification information;

[0070] S4: Based on the damage type classification information, obtain the traffic flow data and vehicle axle load data of the highway section corresponding to the damaged area, extract the average axle load value, passing frequency and the corresponding frequency value of the highest axle load within each time period, call the depth and area indexes of the damaged area, and conduct an expansion trend evaluation according to the cumulative action weight value of vehicle load impact on each damaged structure to generate damage expansion risk information;

[0071] S5: Call the damage expansion risk information, combine the lane traffic flow proportion, speed limit section and accident record road section number, extract the accident interference intensity coefficient and traffic obstruction penalty factor of the corresponding road section, and combine the area index of the damage type and the structural type risk coefficient to conduct weighted calculation, evaluate the processing priority of road damage, sort the processing order, and obtain the road damage processing sorting information.

[0072] The image clarity screening results include the image path index number, image clarity evaluation value, quality determination status and image validity flag. The structural feature information of the damaged area is specifically the edge contour coordinate group, area closure state, pixel gradient distribution and structural boundary strength. The damage type classification information includes the actual crack area, structural boundary length, damage form number and type attribution label. The damage expansion risk information specifically refers to the load influence parameter, passing frequency coefficient, structural expansion trend value and risk level factor. The road damage processing sorting information includes the priority score value, road section identification number, obstruction weight parameter and sorting output index number.

[0073] Please refer to Figure 2 , and the specific steps for obtaining the image clarity screening results are as follows:

[0074] S111: Obtain the collected road surface image, extract the pixel matrices of the rut area, joint area and edge area in the image, call the image pixel values to construct a gray value distribution histogram, and synchronously extract the noise distribution density value at the gray level mutation position in the image signal to obtain the basic pixel characteristic data;

[0075] Obtain the collected road surface image, based on the image data captured by the imaging device, extract the pixel matrices of the rut area, joint area and edge area, and construct a gray value distribution histogram for each area at 256 gray levels, where the gray value of each pixel point is denoted as P i,j, representing the brightness level value of the j-th pixel in the i-th image. Taking the rut area as an example, if an image with a width of 2048 pixels and a height of 1024 pixels is extracted, the size of the pixel matrix of the entire image is 2048×1024. If the rut area is located in the strip area within 100 pixels on both sides of the central axis in the image, the extraction range of its pixel matrix can be set to all pixel points within columns [974, 1074]. After statistically analyzing its gray-scale distribution, a histogram is plotted with the gray-scale value interval as the horizontal axis and the frequency as the vertical axis. Further, the noise distribution density value at the position of the gray-scale mutation in the image signal is synchronously extracted. The noise density value N i,j can be obtained by detecting the degree of variation of the gray-scale value in a small-range neighborhood in the local image area. For example, taking the pixels in a 5×5 window as a unit, the variance of the gray-scale value is calculated. If the variance is greater than the set noise fluctuation threshold of 25, the pixel neighborhood is marked as a noise area. The setting of this threshold of 25 is based on the typical fluctuation range of the gray-scale difference under non-boundary conditions and is obtained through image comparison experiments under the condition of a neutral gray background. The change trend of the noise comparison benchmark is mainly affected by the overall image mean brightness level and the boundary density distribution. Therefore, this value can vary between 12 and 30 according to the overall gray-scale variance σ of the image, and the median value of 25 is selected as the standard demarcation point. Then, the proportion of the number of noise area pixels in the entire image is statistically calculated as the noise density index. If there are 16000 noise area pixels in the image and the total number of pixels in the image is 2097152, then the noise distribution density value is 16000 / 2097152≈0.0076. After the gray-scale histogram and noise density are extracted, the basic pixel characteristic data for each region of the image is formed, including the pixel distribution trend and noise level of each region, and finally the basic pixel characteristic data is obtained.

[0076] S112: Based on the basic pixel characteristic data, call the degree of change of the gray-scale gradient in the image boundary area and the edge response value of the edge pixel points to construct a gradient intensity equilibrium vector group and an edge response value group, and use the formula:

[0077]

[0078] Perform operations to obtain the clarity evaluation index of the image, and make a position judgment on the clarity evaluation index and the set image quality reference value interval, screen out the image samples that exceed the lower limit of the reference value, and obtain the image validity determination information;

[0079] Among them, S i represents the clarity evaluation index of the i-th image, G i,j represents the gray-scale gradient intensity value of the j-th region in the i-th image, E i,j represents the edge response intensity value of the j-th region in the i-th image, N i,jIt represents the noise density value in the j-th region of the i-th image, and n represents the total number of regions into which the image is divided;

[0080] Based on the basic data of pixel characteristics, the degree of gray gradient change and edge response value in the boundary region are called, and the image sharpness index is calculated. Among them, the gray gradient change value G i,j is obtained by the Sobel operator or the image gray difference method. For example, the horizontal and vertical gray differences in the j-th region are Δx = 12 and Δy = 15 respectively, then The edge response value E i,j is determined by the response intensity of the edge detection operator. For example, based on the Canny edge detection, the edge intensity in the j-th region is 21, and the noise density value N i,j is 0.0076. The sharpness is calculated by substituting all parameters into the following formula:

[0081]

[0082] Suppose the i-th image is divided into 4 regions, and the corresponding parameters are as follows:

[0083] G i,1 = 19.2, E i,1 = 21, N i,1 = 0.0076;

[0084] G i,2 = 23.5, E i,2 = 20, N i,2 = 0.0052;

[0085] G i,3 = 21.1, E i,3 = 19.5, N i,3 = 0.0061;

[0086] G i,4 = 18.6, E i,4 = 17, N i,4 = 0.0080;

[0087] Perform the calculation:

[0088] The numerator part:

[0089] |19.2 - 21| + |23.5 - 20| + |21.1 - 19.5| + |18.6 - 17| = 1.8 + 3.5 + 1.6 + 1.6 = 8.5;

[0090] The denominator part:

[0091]

[0092]

[0093] Clarity index:

[0094] S i = 8.5 / 41.3969 ≈ 0.2054;

[0095] Perform a position judgment on the clarity index and the lower limit 0.18 of the image quality reference value range. The setting of this lower limit is based on the minimum boundary response score and the average contrast fluctuation range of the clearly recognizable edge regions in multiple image samples, and is determined after analyzing the cross-average distribution of the gray-scale gradient intensity and the edge response value. Take 0.18 as the available reference value for image recognition quality. This reference value is mainly controlled by the amplitude of gray-scale change and the average edge response. In batch testing, when S i < 0.18, the image boundary is blurred and the feature structure is incomplete. Therefore, 0.18 is the minimum acceptable limit. Images with a clarity evaluation value greater than this reference are judged as valid images. The current image sample has a clarity index of 0.2054, exceeding the lower limit, indicating that the image quality meets the screening criteria, and generate image validity determination information. This result shows that the image sample has certain usability in terms of edge response and noise control.

[0096] S113: According to the image validity determination information, extract the image paths and identification numbers that meet the image quality reference interval conditions, and set quality level labels for the image set to establish an image clarity screening result;

[0097] According to the image validity determination information, identify the samples in all images whose clarity index meets the set range of the interval, and extract the corresponding image path information and the identification numbers registered in the image database. For example, the path information can be expressed as " / data / road_images / img_0457.jpg", and the number is "IMG0457". At the same time, set quality level labels for the image set according to the image clarity value partition. For example, set the clarity index interval as [0.18, 0.22] and label it as the "medium" level,

[0098] [0.22, 0.25] is the "upper medium" level. This quality classification label is set according to the distribution law of the image clarity index in historical data, and the interval within ±1 standard deviation of the distribution mean of the actual image samples is used for classification zoning to ensure that the screening result matches the edge resolvability performance. Combining the sample clarity value of 0.2054, classify it as an image of the "medium" level, and store this information in the image clarity database to form a three-item combined structure of image path, identification number, and quality level, and establish an image clarity screening result.

[0099] Please refer to Figure 3 , and the specific steps for obtaining the structural feature information of the damaged area are as follows:

[0100] S211: Based on the image clarity screening results, extract the pixel matrices of the rut edge area, the central partition area, and the staggered joint area in the screened images, locate the horizontal and vertical pixel distribution position indexes within each area, calculate the gray difference sequence and the arrangement angle gradient sequence of adjacent pixels, and statistically calculate the gray mean difference and the direction distribution stability of each area to obtain the area gray and direction statistical values;

[0101] Based on the image clarity screening results, first extract the image index information located in the rut edge area, the central partition area, and the staggered joint area in the screened image set, and locate its spatial position in the image matrix. Then, based on the pixel matrix corresponding to each image, establish a two-dimensional gray distribution array, and respectively determine the horizontal and vertical pixel arrangement directions for different areas. On this basis, extract the gray values between any two adjacent pixel points in each row and each column respectively to construct a gray difference sequence. Set the sample image number as I1. Then, randomly select the 5th row and the 6th row in the central partition area and extract the nth pixel values as 145 and 130 respectively. The gray difference can be obtained as |145 - 130| = 15. After performing this process on all pixel points in the same area, a complete gray difference sequence is obtained. At the same time, taking each row or each column as a unit, judge the direction of gray value change, calculate its angle difference, and form an arrangement angle gradient sequence. Among them, when the gray change direction difference between two adjacent rows is 10 degrees and the directions are the same, it can be considered that the direction distribution is stable. After statistically calculating all pixel direction differences in this area, calculate the standard deviation, and use less than 5 degrees as the direction distribution stability judgment standard. If the standard deviation is 4.3 degrees, it is determined to be stable. Furthermore, statistically calculate the gray mean and calculate the difference between it and the gray mean of the entire image. In the example, if the area gray mean is 123 and the entire image gray mean is 110, the mean difference is 13. Record the area gray mean difference and the direction distribution standard deviation together as the area gray and direction statistical values.

[0102] S212: According to the area gray and direction statistical values, extract the gray mutation gradient value, the contrast enhancement amplitude value, and the boundary line angle offset value of each structural boundary line segment in the image, perform a boundary response reconstruction operation on the edge of each segment, and at the same time perform a local brightness increment operation on the pixel group with brightness lower than the median value of the global image brightness. Use the formula:

[0103]

[0104] Perform operations to obtain the structure offset response coefficient;

[0105] where, R s represents the structure offset response coefficient, D k represents the normalized value of the gray mutation gradient of the kth boundary line, C k represents the normalized amplitude value of the contrast enhancement of the kth segment, θ kRepresents the normalized value of the angular offset of the k-th boundary line, M k Represents the normalized value of the brightness increment of the k-th boundary line, α is the contrast adjustment coefficient for adjusting C k The influence on the overall result, β is the offset suppression factor for adjusting θ k The weight of the role in the boundary change, N R Is the number of structural boundary line segments in the image;

[0106] According to the regional gray level and direction statistical values, extract the information of the structural boundary line segments in the image, and perform the gray level mutation gradient extraction process on each structural boundary line segment one by one to obtain the change amount of the breakpoint intensity formed by the rapid increase of the gray level value within any paragraph. Assume that the pixel values before and after the gray level mutation point of a boundary line segment in the image are 90 and 150 respectively, then the gray level mutation gradient is

[0107] |150 - 90| = 60, and this value is denoted as D after normalization k = 0.6. Synchronously extract the contrast enhancement amplitude value of this line segment, that is, the difference between the maximum and minimum gray level values divided by the total gray level span. Assume that the maximum gray level value is 160 and the minimum gray level value is 100, then the contrast enhancement is (160 - 100) / 255 ≈ 0.235, and after normalization it is C k = 0.235. In addition, call the aforementioned boundary direction information, compare the angle of the current line segment with the main direction angle of the structure. Assume that the offset angle is 12 degrees, and after quantization and normalization, θ k = 0.12. After processing the brightness enhancement of the edge line segment, the measured edge brightness increment is 40, and after normalization it is M k = 0.16. Then set α = 1.5 and β = 0.8 respectively, and substitute the values into the structural offset response coefficient calculation formula:

[0108]

[0109] Assume that the corresponding parameters of the subsequent 4 boundary line segments are:

[0110] D2 = 0.55,, C2 = 0.20, M2 = 0.14, θ2 = 0.10;

[0111] D3 = 0.58, C3 = 0.22, M3 = 0.15, θ3 = 0.09;

[0112] D4 = 0.62, C4 = 0.18, M4 = 0.17, θ4 = 0.13;

[0113] D5 = 0.59, C5 = 0.21, M5 = 0.19, θ5 = 0.11;

[0114] Substitute them in turn to get:

[0115]

[0116] The advantage of the structural offset response coefficient is that through the comprehensive relationship expression among gray mutation, contrast enhancement, and angle offset, a quantitative response to the integrity of the image boundary structure and local perturbation is achieved.

[0117] Among them, D k represents the normalized value of the gray mutation gradient of the k-th segment of the boundary line, which is obtained by normalizing the gray difference between adjacent edge pixels by 255. C k represents the normalized amplitude value of the contrast enhancement of the k-th segment, which is obtained by normalizing the gray scale span of the boundary line segment with respect to the maximum span of the entire region. θ k represents the normalized value of the angle offset of the k-th segment of the boundary line, which is the angle between the current line segment and the main direction of the region and is normalized to the 0-1 interval. M k represents the normalized value of the brightness increment of the k-th segment of the boundary line, which is obtained by normalizing the difference after the average brightness of the image is enhanced. α is the contrast adjustment coefficient, which is set to 1.5 to expand the influence on the boundary difference degree. k β is the offset suppression factor, which is set to 0.8 to balance the weight influence of the angle perturbation of the boundary line. N R is the number of structural boundary line segments in the image, which is set to 5 in this example.

[0118] S213: Call the structural offset response coefficient, extract the pixel line segment group in the boundary structure segment whose closed trend value is greater than the average offset response, calculate the closed arc distance between the endpoints and the intersection frequency of the segments, and judge whether the closed curve forms a continuous closed path according to the curve density distribution, mark the regional contour structure of the formed closed boundary, and establish the structural feature information of the damaged area;

[0119] Call the structural offset response coefficient, extract the pixel segment group with a high closed trend among the boundary line segments whose structural offset response coefficient is greater than its average value, calculate the Euclidean distance between the starting point and the ending point of each line segment respectively, judge its intersection frequency by combining the number of intersection overlaps with adjacent line segments, further analyze the curvature change trend and connection order between the pixel points forming a continuous curve in the closed segment, confirm whether there is a continuous closed path. For example, if the closed trend value reaches 0.65 in a section of the image and the average value of the offset response coefficient is 0.55, then extract this section as the target segment, the distance between the endpoints of the segment is 15.4 pixels, and the intersection frequency is 3 times. Judge that its closed path is continuous, mark the regional contour structure covered by this path, and finally establish the structural feature information of the damaged area.

[0120] Please refer to Figure 4 for the specific steps of obtaining the damage type classification information:

[0121] S311: Based on the structural feature information of the damaged area, extract the edge pixel path corresponding to each area, call the sequence of edge pixel index values, calculate the cumulative value of the Euclidean distance between adjacent pixel points, count the total edge path length, and call the conversion coefficient between the number of pixel units within the corresponding closed contour and the pixel resolution to calculate the actual area of the closed area, and obtain the damaged area and the boundary quantization value;

[0122] Based on the structural feature information of the damaged area, first extract the edge pixel path corresponding to each damaged area in the image, that is, obtain the set of pixel coordinates included in the closed contour. For each pair of consecutive pixel points in each edge path, call the differences in their horizontal and vertical coordinates to construct an Euclidean distance calculation formula, using the formula: Perform operations on all point pairs in the path and accumulate their distances. Suppose there are 200 valid point pairs in the path, then the edge path length is calculated as If the calculation result is 252.6 pixel unit lengths, then next locate the contour boundary of the area enclosed by this edge path, construct a pixel mask area based on the pixel coordinate enclosure structure, count the number of all pixel units within the mask, assume that the total number of pixels within the closed contour of a certain area is 3450 units, and then call the pixel resolution conversion coefficient attached to the road image acquisition device. For example, if the resolution is 0.0045 square meters per pixel, then the actual area of this closed area is calculated as 3450×0.0045 = 15.525 square meters, obtain the area value of the closed area, based on the relationship between the edge path length and the pixel resolution, and at the same time convert the boundary length to 252.6×√(0.0045)≈16.94 meters. Further combine the area value and the boundary length to form a quantitative expression of the damaged area, which is an important basic index for subsequent judgment of the damage type, and obtain the damaged area and the boundary quantization value.

[0123] S312: According to the damaged area and the boundary quantization value, extract the total edge length value, the closed area value and the contour average curvature value of each damaged area, call the pre-set damage form feature matching template, using the formula:

[0124]

[0125] Perform operations to obtain the structural form matching difference value of the damaged area to be classified;

[0126] Among them, S z represents the structural form matching difference value, A d is the normalized area value of the area to be recognized, A m is the standard area value of the matching template, L d is the boundary length value of the area to be recognized, L m is the boundary length value of the matching template, C d is the average curvature value of the area to be recognized, C mis the reference curvature value for matching the template, and γ is the curvature difference adjustment coefficient;

[0127] According to the damaged area and the boundary quantization value, extract the total edge length value, the enclosed area value, and the contour average curvature value of each damaged area. The average curvature value is obtained by performing the radian difference method on each arc segment in the boundary path, that is, calculating the angular offset value of the middle point for every three consecutive points. Assuming that the regional boundary consists of 120 arc segments, the corresponding curvature array C is obtained d , and then call the preset damaged form matching template database. The template records the standard area value, the boundary length value, and the average curvature value of various common damaged types. For example, for the pothole type template, its standard area value A m = 12.00 square meters, the standard boundary length L m = 14.5 meters, and the average curvature C m = 0.06. Assume that the actually extracted area A d = 15.525 square meters, the boundary length L d = 16.94 meters, and the curvature C d = 0.055. Substitute into the formula:

[0128]

[0129] where the area difference is 0.29375, the boundary difference is 0.16862, and the curvature difference is 0.005. If the curvature difference adjustment coefficient γ = 4.0 (this value is set with reference to the experience of the image edge radian sensitivity. The basis is that at the sampling accuracy of the high-precision image resolution of 0.0045 square meters / pixel, the boundary curvature difference needs to be magnified four times to highlight the contour structure difference. If the actual value of the curvature difference is less than 0.01, the matching and discrimination ability will decline, and the set value should be controlled within the range of [2, 6]), then finally:

[0130] S z = 0.29375 + 0.16862 + 4.0·0.005 = 0.29375 + 0.16862 + 0.02 = 0.48237;

[0131] That is the structural form matching difference value of the damaged area under this template, which is used for subsequent minimum difference matching operations.

[0132] S313: Call the structural form matching difference value of the damaged area to be classified, extract the template label corresponding to the minimum structural form matching difference value, identify the corresponding damaged type, screen the set of numbers with difference values less than the structural form tolerance threshold in the damaged area, establish the type index mapping relationship and generate the annotation value, and establish the damaged type classification information;

[0133] Call the structural form matching difference value, extract all template numbers that match the current damaged area from the preset damaged templates, and calculate S for all templates z Sort them, extract the template number with the smallest value as the optimal matching type of the current damaged area, record the corresponding template label as the identified damaged type number, and set a structural form tolerance threshold in subsequent processing. For example, set it to 0.55 (this value refers to the minimum recognition critical value in the similar matching process of multiple damaged types. If the difference value is less than this value, it can be considered a reliable match; if it is greater than this value, further classification is required). Then extract the set of numbers with all S z values less than 0.55 to form a candidate type set, construct an index mapping of this damaged area in the damaged type dimension, record and manage the mapping results using type labels, and finally obtain the damaged type classification information corresponding to each damaged area.

[0134] Please refer to Figure 5 , and the steps for obtaining the damaged expansion risk information are specifically as follows:

[0135] S411: Based on the damaged type classification information, extract the highway section numbers associated with each damaged area, obtain the traffic flow data and vehicle axle load data obtained by the vehicle detection equipment corresponding to the section during the continuous time period, count the vehicle passing frequencies for each time period, calculate the average axle load value and the maximum axle load occurrence frequency for each time period, and establish a traffic load statistical value;

[0136] Based on the damaged type classification information, extract the highway section numbers associated with each damaged area, match the area with the numbered items recorded in the road database, retrieve the vehicle detection equipment ID attached to the corresponding highway section number and extract its continuously recorded time series data. The extraction period is set to continuous sampling for 24 hours, divided into 96 15-minute time periods. The number of vehicles passing through the detection point in each time period is collected as the passing frequency, the axle load value corresponding to each passing obtained by the vehicle pressure transmission sensor is recorded and integrated according to the time dimension, and the single-axle load data set of all passing vehicles in each time period is obtained. This data set is input into the axle load processing module for processing, and the average value and maximum value calculation operations are respectively performed to obtain the vehicle average axle load value and the maximum axle load value for each time period. The repeated occurrence times of each maximum value are counted in the maximum axle load value data set, and used as the corresponding frequency data of the maximum axle load. Finally, a triple data structure for each time period is formed, that is, the average axle load value, the maximum axle load value, and the maximum value frequency, and is recorded in a quadruple form combined with the time period passing frequency to establish a traffic load statistical value

[0137] S412: According to the statistical value of the passing load, call the area quantization value and depth estimation value of the corresponding damaged area, extract the damage type and structure code, match the preset load impact structure action coefficient, perform the structural cumulative impact assessment operation, and use the formula:

[0138]

[0139] Calculate the damage impact response value of each structure in the corresponding time period, and obtain the structural impact response coefficient;

[0140] Among them, R d represents the structural impact response coefficient, H a,t is the average axle load value in the t-th time period, H m,t is the maximum axle load value in the t-th time period, S z,t is the vehicle passing frequency in the t-th time period, A d,t is the area quantization value of the area in the t-th time period, D d,t is the damage depth normalization value in the t-th time period, λ is the depth enhancement factor, and T is the total number of time periods;

[0141] According to the statistical value of the passing load, call the area quantization value and depth estimation value of each damaged area obtained in the previous steps. The area quantization value is calculated based on the total number of pixel units and the pixel conversion coefficient within the closed contour generated in S311. The depth estimation value is estimated according to the pre-image illumination shadow gradient and the historical structural vertical deformation model. Extract the structure code matching the current damage structure type, and call the preset load impact structure action coefficient library in the storage structure. Match the corresponding impact response template according to the structure code, extract the average axle load value, maximum axle load value, and passing frequency from the passing load statistical value in each time period, and combine the area and depth quantization values, and substitute them into the formula:

[0142]

[0143] Perform the structural impact response calculation for each time period, accumulate and sum according to the set number of time periods and perform normalization processing to obtain the structural impact response coefficient. In this embodiment, set T to 96 and λ to 1.2. The example data of a damaged area in the selected test area is as follows: In the 1st time period, H a,1 = 4.5t, H m,1 = 7.2t, S z,1 = 18 times, A d,1 = 2.8m 2 , D d,1 = 0.05m, substitute into the calculation:

[0144]

[0145] Repeat the calculation 96 times and take the arithmetic mean as R d, the final result example is R d = 76.4. The innovation of the formula lies in that by simultaneously introducing the area and depth terms to construct a synthetic radical expression, and superimposing the axle load difference and traffic frequency to form a triple impact term, the structural response evaluation has multi-dimensional index coupling and time resolution. Among them, R d represents the structural impact response coefficient, H a,t is the average axle load value in the t-th time period, H m,t is the maximum axle load value in the t-th time period, S z,t is the traffic frequency in the t-th time period, A d,t is the area quantization value of the region in the t-th time period, D d,t is the normalized damage depth value in the t-th time period, λ is the depth enhancement factor, and T is the total number of time periods

[0146] S413: Call the structural impact response coefficient, perform grade matching on the structural codes, response values, and threshold grade tables corresponding to each damage type, assign expansion risk marker labels and output the number comparison table, and establish damage expansion risk information;

[0147] Call the structural impact response coefficient, establish a key-value pair mapping between the structural codes corresponding to each obtained damage type and their R d values, extract the impact response intervals corresponding to different grades in the risk threshold grade table. For example, the interval for grade I is [0, 20], grade II is (20, 50], grade III is (50, 80], grade IV is (80, 110], and grade V is (110, +∞). Judge the interval where each R d value is located and assign a risk grade. In the example, the R d corresponding to the structural code T1 is 76.4, which falls into the grade III interval. The matching result is the grade III label. At the same time, it forms a record item T1-III with the structural code, and integrates the matching numbers of all structures to establish a number comparison table, forming a risk information record set, and finally establishing damage expansion risk information.

[0148] Please refer to Figure 6 , the steps for obtaining the road damage treatment sorting information are specifically as follows:

[0149] S511: Call the damage expansion risk information, extract the corresponding road section numbers, obtain the lane traffic flow ratio, speed limit section identifier, and historical accident record information corresponding to the road section, match the impact degree grade value in the accident record and combine it with the speed limit section number, calculate the accident interference grade value and convert it into a standardized index, and adjust the penalty factor in combination with the traffic flow ratio to obtain the traffic interference coefficient value;

[0150] Call the damaged expansion risk information, extract the highway section numbers associated with each damaged area. For example, the damaged area numbered A012 corresponds to the highway section numbered R007. Retrieve the lane traffic flow ratio, speed limit section identification number, and accident number set recorded in its traffic monitoring system. Accident numbers are such as R071 and R105. Extract the accident impact level in the historical records item by item and match its standardized value in the level conversion table. For example, level 5 is mapped to 0.9, and level 3 is mapped to 0.5. Select the maximum value B u = 0.9 as the accident impact level index for the current area. Subsequently, extract the lane traffic flow ratio f = 0.78 corresponding to this section, and use this value together with the accident level index to calculate the traffic interference penalty factor. The calculation method of the penalty factor is F u = μ·f·B u , where the traffic penalty normalization coefficient μ = 0.85. Substitute and calculate to get F u = 0.85·0.78·0.9 = 0.5967. This value is used as the standardized traffic interference coefficient value to participate in the subsequent scoring process.

[0151] S512: According to the traffic interference coefficient value, extract the corresponding damaged area structure type code and area normalization value, call the structure type risk factor weight, and perform a comprehensive priority score calculation according to the predefined risk assessment expression. Use the formula:

[0152]

[0153] Perform operations to obtain the weighted priority score value for each damaged area;

[0154] Among them, P u represents the weighted priority score value, R u is the structural risk factor of the area, A u is the area normalization value of the damaged area, F u is the traffic interference coefficient value of the area, B u is the accident impact level index of the area, and η is the accident penalty enhancement factor;

[0155] According to the traffic interference coefficient value F u = 0.5967, extract the structure type code of the current damaged area as S_C06, and the structural risk factor corresponding to this structure type is R u = 2.0. The accident impact level index of the area extracted in the previous step is B u = 0.9. The accident penalty enhancement factor is set to η = 1.2. This value is set according to the road grade division, 1.2 for the main road, 1.0 for the secondary road, and 0.8 for the branch road. Call the area normalization value of the area as A u = 0.315. Substitute the parameters into the priority score formula:

[0156]

[0157] Among them, P u represents the weighted priority score value of the damaged area, and R u represents the structural risk factor, and A u represents the normalized damaged area value, and F u represents the traffic interference coefficient, and B u represents the accident impact level index, and η represents the accident penalty enhancement factor. The scoring result will be used as the sorting basis. The advantage of this formula is that by introducing the accident interference impact index B u and its weighting factor η into the scoring formula, and jointly processing the weighted denominator with the traffic flow penalty term F u it avoids the phenomenon of distorted scoring values in areas with high traffic density or frequent accidents, and strengthens the multi-factor coordinated evaluation ability in the sorting of structural damage repair.

[0158] S513: Based on the weighted priority score value of each damaged area, perform a sorting process on the score values of each damaged area, combine the road section number and the corresponding sorting result to construct a processing queue index table, and establish road damage processing sorting information;

[0159] Based on the weighted priority score value P u of each damaged area, extract the combination list of area numbers and corresponding score values, perform a descending sorting process on the scores. For example, the score value corresponding to the number A012 is 0.2354, A015 corresponds to 0.4121, and A017 corresponds to 0.2933. Arrange them in descending order of scores as A015, A017, A012, establish a one-to-one mapping between the numbers and the sorting index values, such as A015→1, A017→2, A012→3, and finally construct the complete road damage processing sorting information, which is used to call the sorting priority in the subsequent scheduling interface to complete the repair task assignment and order management.

[0160] An intelligent highway damage identification system based on image analysis, which is used to implement the intelligent highway damage identification method based on image analysis. The system includes:

[0161] The image screening and processing module acquires the collected road surface images, calls the gray value distribution histogram of the damaged areas in the images and the overall noise distribution density value of the images, compares the degree of gray gradient change and the boundary pixel edge response value, calculates the image clarity evaluation index, compares it with the set image quality reference value, screens out the images that do not meet the quality reference value conditions, and records the image path and identification number to generate the image clarity screening result;

[0162] Based on the image sharpness screening results, the damage feature recognition module performs boundary localization processing on the pixel point distributions in the rut edge area, the central partition area, and the staggered joint area of the image according to the screened images, obtains the local contrast value and the spatial frequency distribution value in the image of each area, performs edge sharpening operations on the image according to the boundary strength value and the contrast enhancement weight, and performs local brightness adjustment processing on the boundary area, identifies the damaged edge curve, notch shape, and closed contour features in the enhanced image, and generates damage area structure feature information;

[0163] Based on the damage area structure feature information, the damage type evaluation module calls the pixel resolution coefficient of the road image according to the edge length, curvature range, and closed area area value of the damage area, converts the area and boundary length of each damage area in the actual road surface, and makes a matching judgment based on the structural form of the damage and the known boundary form of the damage type to identify the damage type and obtain the damage type classification information;

[0164] Based on the damage type classification information, the damage expansion analysis module obtains the traffic flow data and vehicle axle load data of the highway section corresponding to the damage area, extracts the average vehicle axle load value, passing frequency, and the corresponding frequency value of the highest axle load in each time period, calls the depth and area indicators of the damage area, and evaluates the expansion trend according to the cumulative action weight value of the vehicle load impact on each damage structure to generate damage expansion risk information;

[0165] The damage treatment evaluation module calls the damage expansion risk information, combines the lane traffic flow ratio, speed limit section, and accident record section number, extracts the accident interference intensity coefficient and traffic obstruction penalty factor of the corresponding road section, combines the area indicator of the damage type and the structural type risk coefficient, performs weighted calculation, evaluates the treatment priority of the road damage, and sorts the treatment order to obtain the road damage treatment sorting information.

[0166] The above is only the preferred embodiment of the present invention, and it is not intended to limit the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution content of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. An intelligent recognition method for highway damage based on image analysis, characterized in that, It includes the following steps: S1: Obtain the collected road surface images, calculate the image clarity evaluation index, screen out the images that do not meet the quality reference value conditions, record the image paths and identification numbers, and generate the image clarity screening results; S2: Based on the image clarity screening results, obtain the local contrast value and spatial frequency distribution value in each regional image, perform edge sharpening operations on the images according to the boundary intensity value and contrast enhancement weight, identify the damaged edge curves, notch shapes and closed contour features in the enhanced images, and generate the damaged area structure feature information; S3: According to the damaged area structure feature information, call the pixel resolution coefficient of the road image, convert the area and boundary length of each damaged area in the actual road surface, and perform matching judgment based on the damaged structural form and the known damaged type boundary form to identify the damaged type and obtain the damaged type classification information; S4: Based on the damaged type classification information, extract the average axle load value, passing frequency and the corresponding frequency value of the highest axle load within each time period, and perform an expansion trend evaluation according to the cumulative action weight value of the vehicle load impact on each damaged structure to generate the damaged expansion risk information.

2. The intelligent road damage recognition method based on image analysis according to claim 1, wherein, The image clarity screening results include the image path index number, image clarity evaluation value, quality determination status and image validity flag. The damaged area structure feature information is specifically the edge contour coordinate group, area closed state, pixel gradient distribution and structural boundary intensity. The damaged type classification information includes the actual crack area, structural boundary length, damaged form number, and type attribution label. The damaged expansion risk information specifically refers to the load impact parameter, passing frequency coefficient, structural expansion trend value and risk level factor.

3. The intelligent road damage identification method based on image analysis according to claim 2, wherein The specific steps for obtaining the image clarity screening results are as follows: S111: Obtain the collected road surface images, extract the pixel matrices of the rut area, joint area and edge area in the images, call the image pixel values to construct a grayscale value distribution histogram, and simultaneously extract the noise distribution density value at the gray level mutation position in the image signal to obtain the basic pixel characteristic data; S112: Based on the basic pixel characteristic data, call the gray level gradient change degree and edge response value of the image boundary area to construct a gradient intensity equalization vector group and an edge response value group, and use the formula: Operate to obtain the image clarity evaluation index, and perform a position judgment on the clarity evaluation index and the set image quality reference value interval, screen out the image samples that exceed the lower limit of the reference value, and obtain the image validity determination information; Among them, S i represents the clarity evaluation index of the i-th image, G i,j represents the gray gradient intensity value of the j-th region in the i-th image, E i,j represents the edge response intensity value of the j-th region in the i-th image, N i,j represents the noise density value within the j-th region in the i-th image, and n represents the total number of regions into which the image is divided; S113: According to the image validity determination information, extract the image paths and identification numbers that meet the image quality reference interval conditions, and set quality level labels for the image set to establish the image clarity screening results.

4. The intelligent road damage identification method based on image analysis according to claim 3, characterized in that The specific steps for obtaining the damaged area structure feature information are as follows: S211: Based on the image sharpness screening results, extract the pixel matrices of the rut edge region, the central partition region, and the staggered joint region in the screened images, locate the horizontal and vertical pixel distribution position indexes within each region, calculate the adjacent pixel gray difference sequence and the arrangement angle gradient sequence, and statistically calculate the gray mean difference and the direction distribution stability of each region to obtain the region gray and direction statistical values; S212: According to the region gray and direction statistical values, extract the gray mutation gradient value, the contrast enhancement amplitude value, and the boundary line angle deviation value of each structural boundary line segment in the image, perform a boundary response reconstruction operation on each line segment edge, and at the same time perform a local brightness increment operation on the pixel group with a brightness lower than the median value of the overall image brightness. Use the formula: Calculate to obtain the structural offset response coefficient; Among them, R s represents the structural offset response coefficient, D k represents the normalized value of the gray-scale mutation gradient of the k-th segment boundary line, C k represents the normalized amplitude value of the contrast enhancement of the k-th segment, θ k represents the normalized value of the angular offset of the k-th segment boundary line, M k represents the normalized value of the brightness increment of the k-th segment boundary line, α is the contrast adjustment coefficient, β is the offset suppression factor, N R is the number of structural boundary line segments in the image; S213: Call the structural offset response coefficient, extract the pixel line segment group with a closed trend value greater than the offset response mean in the boundary structure segment, calculate the closed arc distance between the endpoints and the segment intersection frequency, determine whether the closed curve forms a continuous closed path according to the curve density distribution, label the area contour structure of the formed closed boundary, and establish the damaged area structure feature information.

5. The intelligent recognition method for highway damage based on image analysis according to claim 4, characterized in that The specific steps for obtaining the damage type classification information are as follows: S311: Based on the damaged area structure feature information, extract the edge pixel path corresponding to each region, call the edge pixel index value sequence, calculate the cumulative value of the Euclidean distances between adjacent pixel points, statistically calculate the total edge path length, and call the number of pixel units within the corresponding closed contour and the pixel resolution conversion coefficient to calculate the actual area of the closed region to obtain the damaged area and the boundary quantization value; S312: According to the damaged area and the boundary quantization value, extract the total edge length value, the closed region area value, and the contour average curvature value of each damaged region, call the preset damage form feature matching template, and use the formula: Calculate to obtain the structural form matching difference value of the damaged region to be classified; Among them, S z represents the structural form matching difference value, A d is the normalized area value of the area to be recognized, A m is the standard area value of the matching template, L d is the boundary length value of the area to be recognized, L m is the boundary length value of the matching template, C d is the average curvature value of the area to be recognized, C m is the reference curvature value of the matching template, and γ is the curvature difference adjustment coefficient; S313: Call the structural form matching difference value of the damaged region to be classified, extract the template label corresponding to the minimum structural form matching difference value, identify the corresponding damage type, screen the set of numbers with a difference value less than the structural form tolerance threshold in the damaged region, establish a type index mapping relationship and generate a marked value, and establish the damage type classification information.

6. The intelligent recognition method for highway damage based on image analysis according to claim 5, characterized in that, The specific steps for obtaining the damage expansion risk information are as follows: S411: Based on the damage type classification information, extract the highway section numbers associated with each damaged region, obtain the traffic flow data and vehicle axle load data within a continuous time period obtained by the vehicle detection equipment corresponding to the section, count the vehicle passing frequencies within each time period, calculate the average axle load value and the maximum axle load occurrence frequency within each time period, and establish the passing load statistical value; S412: According to the passing load statistical value, call the area quantization value and the depth estimation value of the corresponding damaged region, extract the damage type and the structure code, match the preset load impact structure action coefficient, and perform a structural cumulative impact assessment operation. Use the formula: Calculate the damage impact response value corresponding to each time period under each type of structure to obtain the structural impact response coefficient; Among them, R d represents the structural impact response coefficient, H a,t is the average axle load value in the t-th time period, H m,t is the maximum axle load value in the t-th time period, S z,t is the vehicle passing frequency in the t-th time period, A d,t is the regional area quantization value in the t-th time period, D d,t is the normalized damage depth value in the t-th time period, λ is the depth enhancement factor, and T is the total number of time periods; S413: Call the structural impact response coefficient, perform level matching on the structural codes, response values, and threshold level tables corresponding to each damage type, assign expansion risk marker tags, output a number comparison table, and establish damage expansion risk information.

7. The intelligent road damage recognition method based on image analysis according to claim 6, wherein, The method further includes: S5: Call the damage expansion risk information, combine the lane traffic flow ratio, speed limit section, and accident record section number, evaluate the processing priority of road damage, sort the processing order, and obtain road damage processing sorting information; The road damage processing sorting information includes a priority score value, a road section identification number, an obstacle weight parameter, and a sorting output index number.

8. The intelligent road damage identification method based on image analysis according to claim 7, wherein The specific steps for obtaining the road damage processing sorting information are as follows: S511: Call the damage expansion risk information, extract the corresponding section number, obtain the lane traffic flow ratio, speed limit section identifier, and historical accident record information corresponding to the section, match the impact degree level value in the accident record, combine with the speed limit section number, calculate the accident interference level value, convert it into a standardized index, and adjust the penalty factor in combination with the traffic flow ratio to obtain the traffic interference coefficient value; S512: According to the traffic interference coefficient value, extract the corresponding damaged area structure type code and area normalization value, call the structure type risk factor weight, and perform a comprehensive priority score calculation according to the predefined risk assessment expression, using the formula: Operate to obtain the weighted priority score value of each damaged area; Among them, P u represents the weighted priority score value, R u is the structural risk factor of the area, A u is the normalized value of the damaged area of the area, F u is the traffic interference coefficient value of the area, B u is the accident impact level index of the area, and η is the accident penalty enhancement factor; S513: Based on the weighted priority score value of each damaged area, perform a score value sorting process on each damaged area, combine the section number and the corresponding sorting result to construct a processing queue index table, and establish road damage processing sorting information.

9. An intelligent road damage recognition system based on image analysis, characterized in that, The system is used to implement the intelligent road damage identification method based on image analysis according to any one of claims 1-8. The system includes: The image screening and processing module acquires the collected road surface images, calculates the image clarity evaluation index, screens out the images that do not meet the quality reference value conditions, records the image path and identification number, and generates an image clarity screening result; The damage feature recognition module, based on the image clarity screening result, obtains the local contrast value and spatial frequency distribution value in each area image, performs edge sharpening operations on the image according to the boundary strength value and contrast enhancement weight, and identifies the damage edge curve, notch shape, and closed contour features in the enhanced image to generate damage area structure feature information; The damage type evaluation module, according to the damage area structure feature information, converts the area and boundary length of each damaged area in the actual road surface, and performs a matching judgment based on the damaged structure form and the known damage type boundary form to identify the damage type and obtain the damage type classification information; The damage expansion analysis module, based on the damage type classification information, evaluates the expansion trend according to the cumulative action weight value of vehicle load impact on each damaged structure, and generates damage expansion risk information; The damage handling evaluation module calls the damage expansion risk information, combines the lane traffic flow ratio, speed limit section, and accident record section number, evaluates the handling priority of road damage, sorts the handling order, and obtains the road damage handling sorting information.

Images