A machine recognition-based road quality automatic intelligent detection method and system

Abstract

The application discloses a kind of based on machine identification's road quality automation intelligent detection method and system, it is related to road detection technical field, the detection method steps include, according to the original image of captured road surface, after executing gray normalization operation, standard image matrix is constructed;After executing directional gray difference value calculation, build exception response matrix, using eight neighborhood connectivity determination method, candidate connected region is screened out, the area of each candidate connected region is compared and analyzed, and abnormal crack region is screened out, and crack region feature data is constructed;In combination with inverse tangent function, the surface structure response degree of each crack connected region is analyzed, the structure response coefficient of each crack connected region is determined, in combination with the abnormal texture response value of each candidate pixel point in each crack connected region extracted, determine the road surface abnormal intensity index;Judge whether the current road surface structure exists exception, and issue corresponding grade quality detection result.

Description

An automated intelligent detection method and system for road quality based on machine recognition Technical Field

[0001] This invention relates to the field of road inspection technology, specifically to an automated intelligent inspection method and system for road quality based on machine recognition. Background Technology

[0002] In the maintenance of urban roads, highways, and roads in urban-rural fringe areas, the structural integrity of the road surface directly affects driving safety and passenger comfort. To ensure the stability of traffic operations and the timeliness of road maintenance, real-time monitoring of the road surface condition and accurate identification of fault areas have become core aspects of road maintenance management. Unlike traditional inspection methods that rely on manual inspection and recording, automated inspection solutions based on machine vision, image recognition, and structural modeling algorithms have been gradually developed in recent years, enabling high-frequency inspections of cracks and other abnormal conditions.

[0003] While some image acquisition and visual recognition technologies have been introduced into existing road inspection and quality assessment processes, a series of insurmountable limitations remain in the core aspects of automatic crack region extraction, quantitative modeling of structural response intensity, anomaly index calculation, and connectivity region discrimination. Image grayscale is affected by external lighting and road surface reflection, making it difficult to achieve consistent and normalized image feature processing, resulting in blurred grayscale recognition boundaries in crack regions and low accuracy in anomaly region extraction. Furthermore, existing methods rely solely on simple edge detection and pixel threshold judgment, making it difficult to perform coherent spatial structure analysis of crack connectivity regions, thus hindering the effective differentiation between cracks and noise areas. The lack of a structural response modeling mechanism for crack regions makes it difficult to analyze crack configuration from the perspectives of displacement difference, area density, and principal axis length. These problems limit the real-time performance, stability, and automation level of road condition determination, making it difficult for traditional image recognition systems to meet the management needs of large-scale, multi-scenario, and structurally diverse road systems in road quality inspection.

[0004] The core reason for the above problems stems from the disconnect between image data processing and structural modeling. In actual detection scenarios, the grayscale characteristics, texture continuity, and structural perturbation features of images exhibit strong spatial nonlinearity. If these characteristics are not fully considered in image normalization, crack connectivity analysis, and response index modeling, it will directly lead to a decrease in the accuracy of crack region identification and distortion of structural response representation. Especially in the context of strong light interference or reflective textures, traditional algorithms struggle to separate the feature overlap between normal textures and real cracks, resulting in misidentification of candidate regions, failure to remove noise interference, and strong response distortion. The abnormal consequences of statistical deviations: When cracked areas are incorrectly classified as normal areas, it will cause delays in maintenance and management response, creating potential hazards of road function failure and traffic safety accidents; conversely, if normal textures are misjudged as cracked areas, it will also lead to wasted resource scheduling and false alarms in the detection system. The above phenomena indicate that the existing detection mechanism has not yet built a closed-loop, parametric modeling process from image acquisition, abnormal area identification, structural response calculation to abnormal index quantification. There is an urgent need for an intelligent detection technology path that can support multi-stage analysis and has structural response modeling capabilities to achieve the goal of comprehensive, accurate and automated road quality identification. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an automated intelligent detection method and system for road quality based on machine recognition, which solves the problems mentioned in the background.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an automated intelligent detection method for road quality based on machine recognition, comprising the following steps:

[0007] S1. Based on the original road surface images captured by the camera module fixed on the front bracket of the road inspection vehicle, determine the image feature data information, perform grayscale normalization, and construct a standard image matrix.

[0008] S2. Perform directional gray-level difference calculation on the standard image matrix, and construct an abnormal response matrix by combining the crack prediction response function. Use the eight-neighbor connectivity determination method to screen out candidate connected regions. Compare and analyze the area of ​​each candidate connected region to screen out abnormal crack regions and construct crack region feature data.

[0009] S3. Based on the feature data of the crack area and combined with the arctangent function, analyze the surface structure response degree of each crack connected area, determine the structure response coefficient of each crack connected area, and combine the abnormal texture response value of each candidate pixel in each crack connected area to determine the road surface abnormal intensity index.

[0010] S4. Compare and analyze the pavement anomaly strength index with the preset anomaly strength threshold to determine whether there is anomaly in the current pavement structure, and issue the corresponding level of quality inspection results.

[0011] Preferably, step S1 specifically includes:

[0012] S11. Based on the original road surface images captured by the camera module fixed to the front bracket of the road inspection vehicle, determine the image feature data information, specifically including:

[0013] During the current road surface image acquisition process, the camera module fixed on the front bracket of the road inspection vehicle is triggered by a preset periodic control signal to capture the current road surface image. After grayscale processing, the original road surface image is constructed. The latitude and longitude location information of the current road is located in real time through the GNSS receiving unit, and the latitude and longitude location information of the current road and the original road surface image are sent to the local data center for storage.

[0014] Based on the pixel resolution of the original road surface image, the original road surface image is divided into pixels, and a two-dimensional pixel coordinate plane is established in combination with the image size. At the same time, the gray value and corresponding coordinate position of each pixel in the two-dimensional pixel coordinate plane are recorded to construct image feature data information.

[0015] In the two-dimensional pixel coordinate plane, the origin is the pixel at the bottom left corner of the original image of the road surface, denoted as (0,0). The x-axis extends from left to right, with a value range of x∈[0,W-1], and the y-axis extends from bottom to top, with a value range of y∈[0,H-1]. Here, W represents the number of pixels on the x-axis, and H represents the number of pixels on the y-axis.

[0016] Preferably, step S1 further includes:

[0017] S12. Perform grayscale normalization on the image feature data to construct a standard image matrix, specifically including:

[0018] The gray values ​​of each pixel in the image feature data are extracted, and after pixel statistical processing, the mean and standard deviation of the gray values ​​of all pixels are calculated.

[0019] Based on the mean and standard deviation of the gray values ​​of all pixels, a normalization calculation is performed on the gray values ​​of each pixel in the image feature data to construct a standard image matrix, specifically: In the formula, Fim x,y H represents the standard image matrix. x,yσ represents the gray value of the corresponding pixel, μ represents the mean of the gray values ​​of all pixels, σ represents the standard deviation of the gray values ​​of all pixels, x represents the horizontal coordinate of the corresponding pixel, and y represents the vertical coordinate of the corresponding pixel.

[0020] Preferably, step S2 specifically includes:

[0021] S21. Perform directional gray-level difference calculation on the standard image matrix. Perform first-order derivative operations along the x-axis and y-axis respectively to extract the horizontal and vertical gray-level change rates. The specific extraction process includes:

[0022] Centered on each pixel, a 3×3 pixel matrix around the corresponding pixel is selected as the corresponding local pixel region. The gradient response of the corresponding pixel in the horizontal and vertical directions is extracted by the Sobel operator, and the horizontal gray-scale change rate and vertical gray-scale change rate of each pixel are determined respectively.

[0023] The horizontal and vertical grayscale change rates of each pixel are obtained using the following formulas:

[0024] G x =Fim x+1,y -Fim x-1,y G y =Fim x,y+1 -Fim x,y-1 ;

[0025] In the formula, G x With G y These represent the horizontal and vertical grayscale change rates of the corresponding pixels, respectively.

[0026] S22. Based on the extracted horizontal and vertical grayscale change rates, and combined with the crack prediction response function, after dimensionless processing, the grayscale gradient change amplitude of each pixel is analyzed to construct an abnormal response matrix, specifically: Afm x,y =|G x -G y |;In the formula, Afm x,y G represents the abnormal response matrix. x With G y These represent the horizontal and vertical grayscale change rates of the corresponding pixels, respectively. The abnormal response matrix includes the abnormal texture response values ​​of the corresponding pixels.

[0027] Preferably, step S2 further includes:

[0028] S23. Compare and analyze the abnormal texture response values ​​of each pixel in the abnormal response matrix with preset response thresholds, and use the eight-neighbor connectivity determination method to select several independent candidate connected regions, specifically including:

[0029] If the abnormal texture response value of the corresponding pixel exceeds the preset response threshold, it indicates that the corresponding pixel is a structural texture break and is marked as a candidate pixel; otherwise, it is marked as a normal pixel.

[0030] Based on the labeled candidate pixels, and using the eight-neighbor connectivity method, all consecutive candidate pixels on the plane are aggregated into several independent candidate connected regions, and the abnormal texture response values ​​and coordinate positions of each candidate pixel in the corresponding candidate connected region are recorded.

[0031] Preferably, step S2 further includes:

[0032] S24. The area of ​​the corresponding candidate connected region is recorded as the number of all candidate pixels in the selected candidate connected region. The perimeter of the corresponding candidate connected region is recorded as the number of candidate pixels counted on the boundary contour of the corresponding candidate connected region. The principal axis length of the corresponding candidate connected region is recorded as the maximum coordinate distance between two candidate pixels in the corresponding candidate connected region.

[0033] S25. Compare and analyze the area of ​​each candidate connected region to screen out abnormal crack regions and construct crack region feature data, specifically including:

[0034] If the area of ​​the corresponding selected connected region exceeds the preset area threshold, it indicates that the corresponding selected connected region is an abnormal crack region. After the region is numbered, the area, perimeter, principal axis length and coordinate position of each candidate pixel of the corresponding crack connected region are recorded to construct crack region feature data.

[0035] If the area of ​​the corresponding selected connected region does not exceed the preset area threshold, it indicates that the corresponding selected connected region is a normal noise region, and the corresponding noise region is removed.

[0036] Preferably, step S3 specifically includes:

[0037] S31. Based on the boundary contours of each crack connected region and combined with the crack region feature data, extract the candidate pixel coordinates of the uppermost, lowermost, leftmost and rightmost edges of the boundary contours of each crack connected region. After calculating the difference in centroid displacement in the vertical and horizontal directions, obtain the vertical centroid displacement difference and the horizontal centroid displacement difference of each crack connected region respectively.

[0038] S32. Correlate the vertical centroid displacement difference with the horizontal centroid displacement difference, the area of ​​the region, and the principal axis length of each crack-connected region. Combine this with the arctangent function to analyze the surface structural response of each crack-connected region and determine the structural response coefficient of each crack-connected region. Specifically: In the formula, Srmi ΔDx represents the structural response coefficient of the corresponding crack-connected region. i and ΔDy i Sr represents the difference in horizontal and vertical centroid displacements of the corresponding crack-connected regions, respectively. i and Lr i These represent the area and principal axis length of the corresponding crack connected region, respectively, and arctan(*) represents the tangent function.

[0039] Preferably, step S3 further includes:

[0040] S33. Extract the abnormal texture response values ​​of each candidate pixel in each crack connected region, and combine them with the structural response coefficient of the corresponding crack connected region to determine the road surface abnormal intensity index, specifically including:

[0041] Based on the abnormal response matrix constructed by S22, the abnormal texture response values ​​of each candidate pixel in each crack connected region are extracted, and the mean abnormal texture response value of each crack connected region is obtained by combining the statistical mean calculation algorithm.

[0042] By correlating the mean of the abnormal texture response in each crack-connected region with the structural response coefficient, the degree of structural anomaly of the current pavement is analyzed, and the pavement anomaly strength index of the current pavement is obtained, specifically: In the formula, Zyc represents the pavement anomaly strength index, and Srm represents the pavement anomaly strength index i Afm represents the structural response coefficient of the corresponding crack-connected region. i represents the mean of the abnormal texture response of the corresponding crack connected region, and n represents the number of crack connected regions.

[0043] Preferably, step S4 specifically includes:

[0044] S41. Compare and analyze the current pavement anomaly strength index with the preset anomaly strength threshold to determine whether there is anomaly in the current pavement structure, and issue the corresponding level of quality inspection results, specifically including:

[0045] If the current pavement abnormality strength index does not exceed the abnormality strength threshold, it indicates that the current pavement structure is normal and a first-level quality monitoring result is issued. At this time, the S1 process is executed to continue collecting the next road surface image.

[0046] If the pavement abnormality strength index exceeds the abnormality strength threshold, it indicates that there is an abnormality in the current pavement structure, and a secondary quality monitoring result is issued. At this time, road maintenance personnel are notified to repair the current road and the latitude and longitude location information of the current road is sent.

[0047] An automated intelligent road quality detection system based on machine recognition includes an image preprocessing module, a feature analysis module, a structure evaluation module, and a quality judgment module.

[0048] The image preprocessing module is used to determine image feature data information based on the original road surface image captured by the camera module fixed on the front bracket of the road inspection vehicle, and to construct a standard image matrix after performing grayscale normalization.

[0049] The feature analysis module is used to perform directional gray-level difference calculation on the standard image matrix, and combined with the crack prediction response function, construct an abnormal response matrix. Using the eight-neighbor connectivity determination method, candidate connected regions are selected, the area of ​​each candidate connected region is compared and analyzed, abnormal crack regions are selected, and crack region feature data is constructed.

[0050] The structural assessment module analyzes the surface structural response of each connected region of cracks based on crack region feature data and combined with arctangent function, determines the structural response coefficient of each connected region of cracks, and determines the road surface anomaly intensity index by combining the extracted abnormal texture response values ​​of each candidate pixel in each connected region of cracks.

[0051] The quality assessment module is used to compare and analyze the pavement anomaly strength index with the preset anomaly strength threshold to determine whether there is anomaly in the current pavement structure and issue the corresponding level of quality inspection results.

[0052] This invention provides an automated intelligent detection method and system for road quality based on machine recognition, which has the following beneficial effects:

[0053] (1) By constructing a multi-stage data processing chain consisting of “image normalization, abnormal response extraction, connected component extraction, structural response calculation and index construction”, a closed-loop analysis mechanism is formed that runs through the entire process from image acquisition to abnormal index output. This effectively solves the problems of lack of continuity between image features and structural modeling, lack of parameter support in response calculation process, and low accuracy of abnormal area identification in existing technologies. This detection method does not rely on manual intervention, but standardizes image grayscale with a unified standard, constructs a clear crack response calculation path, and ensures that each abnormal area has a corresponding response index in structure, thereby realizing an objective, quantitative and comparable expression of the road surface state. In addition, by introducing specific measurable parameters such as area, center of gravity displacement and principal axis distance and constructing a reasonable function expression, it has the ability to stably process multiple types of crack structures, avoids misidentification and misjudgment caused by background disturbance or noise texture, fundamentally improves the engineering applicability, scene transfer capability and index calculation consistency of the road detection process, and is conducive to supporting the needs of automated state discrimination and dynamic scheduling management of large-scale traffic networks.

[0054] (2) During image acquisition, factors such as external lighting conditions, differences in road surface reflection, and changes in camera angle often cause significant differences in image grayscale distribution, leading to instability in traditional edge detection methods for crack boundary identification. To address this challenge, this invention establishes a standardized grayscale normalization process in the image preprocessing stage. It uses the mean and standard deviation of the grayscale values ​​of all pixels in the entire image as the processing benchmark. By performing a linear normalization transformation on each pixel, the grayscale values ​​of the image are normalized to a uniform data range under different environments. The mean and standard deviation involved in this processing are calculated by traversing the pixels, eliminating the need for importing... Arbitrarily adjustable coefficients or empirical factors ensure the objectivity and engineering stability of the calculation results. The normalized image serves as the basic input for subsequent analysis modules, effectively eliminating differences in shooting time, environment, and equipment. This eliminates grayscale interference caused by changes in illumination during the extraction of abnormal regions. By maintaining the image spatial structure and adjusting only the grayscale representation, continuous and stable gradient responses are ensured when subsequent modules extract boundaries and gradients in the spatial direction. This greatly enhances the universal adaptability of image structure representation in different scenarios and provides a unified and reliable grayscale basis for the identification of abnormal regions.

[0055] (3) Existing technologies for crack identification generally suffer from problems such as vague judgment criteria, lack of crack spatial morphology modeling, and lack of parameterized expression of response intensity. This makes it difficult to form a unified and comparable standard index for the final anomaly assessment results, which is insufficient to support subsequent scheduling response. In contrast, this road detection method, based on the identification of anomaly areas, further constructs a clear structural response calculation formula and extracts structural parameters including crack area, principal axis length, and the difference between the centroids in the vertical and horizontal directions as the core variables for response value calculation. The formula does not involve weight structure and normalization factor, but only uses directly measured image physical parameters. For example, the principal axis length is measured by the distance of the longest line segment of the region boundary, and the centroid difference is measured by the center of the region shape. The area is calculated by counting the number of pixels, ensuring that all input quantities are repeatable and verifiable in engineering. After response modeling is completed, an anomaly index is calculated based on the structural response value and texture gradient intensity. The influence of extreme values ​​is suppressed according to the logarithmic function form, and a reasonable anomaly judgment basis with numerical scale is constructed. This allows subsequent modules to directly use the index for strategy ranking, scheduling index formulation, or road condition classification. This significantly improves the bottleneck of existing road detection methods that are difficult to quantify the severity of crack structures and difficult to compare and reuse response results. It fills the problem of missing parameterization path from crack morphology recognition to response calculation to index output, ensuring that the entire detection process has a closed-loop structure in mathematical logic and engineering implementation. Attached Figure Description

[0056] Figure 1 is a schematic diagram of the automated intelligent road quality detection method based on machine recognition according to the present invention.

[0057] Figure 2 is a block diagram of an automated intelligent road quality detection system based on machine recognition according to the present invention.

[0058] Figure 3 is a schematic diagram of the logical thinking of the automated intelligent detection method for road quality based on machine recognition according to the present invention. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Example 1

[0061] Please refer to Figures 1 and 3. This invention provides an automated intelligent detection method for road quality based on machine recognition, comprising the following steps:

[0062] S1. Based on the original road surface images captured by the camera module fixed on the front bracket of the road inspection vehicle, determine the image feature data information, perform grayscale normalization, and construct a standard image matrix.

[0063] S2. Perform directional gray-level difference calculation on the standard image matrix, and construct an abnormal response matrix by combining the crack prediction response function. Use the eight-neighbor connectivity determination method to screen out candidate connected regions. Compare and analyze the area of ​​each candidate connected region to screen out abnormal crack regions and construct crack region feature data.

[0064] S3. Based on the feature data of the crack area and combined with the arctangent function, analyze the surface structure response degree of each crack connected area, determine the structure response coefficient of each crack connected area, and combine the abnormal texture response value of each candidate pixel in each crack connected area to determine the road surface abnormal intensity index.

[0065] S4. Compare and analyze the pavement anomaly strength index with the preset anomaly strength threshold to determine whether there is anomaly in the current pavement structure, and issue the corresponding level of quality inspection results.

[0066] In this embodiment, the automated intelligent road quality detection method based on machine recognition unfolds around four consecutive steps: image preprocessing, anomaly extraction, structural modeling, and result judgment. This forms a closed-loop image analysis and structural response calculation process, significantly overcoming the technical challenges of inconsistent image grayscale, blurred boundary recognition, weak crack connectivity analysis, and lack of structural response modeling in traditional detection methods. By performing grayscale normalization on the original images captured by the inspection vehicle's front-end camera module, a standard image matrix is ​​constructed, solving the problem of image grayscale differences caused by changes in ambient lighting. This provides a unified grayscale basis for subsequent image analysis. By performing directional grayscale difference calculation and constructing a response function, combined with the eight-neighbor connectivity determination method, the ability to identify the structural integrity of crack areas is effectively improved. Simultaneously, area comparison suppresses the false extraction of false crack areas. This method addresses the issues of misidentification of candidate regions and difficulty in eliminating noise interference in existing methods. By modeling the arctangent of crack structural parameters and analyzing crack texture response values, it establishes clear and quantifiable structural response coefficients and anomaly intensity indices, replacing previous methods that relied on experience-based judgment or simple edge detection. This enhances the comparability and numerical representation of crack severity. By comparing the anomaly index with a preset threshold, it achieves automated identification and quality output of structural anomalies, providing real-time executable detection results for road management systems. This avoids the problems of delayed response and large judgment errors in traditional manual inspections. Overall, this method establishes a clear mathematical and parametric relationship between image feature extraction and structural information reconstruction, achieving a seamless connection from image perception to structural representation. This makes the road crack identification process highly stable, accurate, and practical for engineering applications.

[0067] Example 2

[0068] Please refer to Figures 1 and 3. Specifically, the steps in S1 include:

[0069] S11. Based on the original road surface images captured by the camera module fixed to the front bracket of the road inspection vehicle, determine the image feature data information, specifically including:

[0070] During the current road surface image acquisition process, the camera module fixed on the front bracket of the road inspection vehicle is triggered by a preset periodic control signal to capture the current road surface image. After grayscale processing, the original road surface image is constructed. The latitude and longitude location information of the current road is located in real time through the GNSS receiving unit, and the latitude and longitude location information of the current road and the original road surface image are sent to the local data center for storage.

[0071] The GNSS receiving unit determines the longitude and latitude of the current road location in real time by receiving positioning signals sent by the global satellite navigation network. The GNSS receiving unit is integrated with the road inspection vehicle to accurately and synchronously record the geographical location information corresponding to the original image of the road surface while the image is being acquired. This binding of spatial data and image data ensures that each image has a clear geographical reference, thereby providing basic data support for subsequent location of abnormal road areas, tracing of structural damage, and operation and maintenance scheduling.

[0072] Based on the pixel resolution of the original road surface image, the original road surface image is divided into pixels, and a two-dimensional pixel coordinate plane is established in combination with the image size. At the same time, the gray value and corresponding coordinate position of each pixel in the two-dimensional pixel coordinate plane are recorded to construct image feature data information.

[0073] In the two-dimensional pixel coordinate plane, the origin is the pixel at the bottom left corner of the original image of the road surface, denoted as (0,0). The x-axis extends from left to right, with a value range of x∈[0,W-1], and the y-axis extends from bottom to top, with a value range of y∈[0,H-1]. Here, W represents the number of pixels on the x-axis, and H represents the number of pixels on the y-axis.

[0074] Specifically, the S1 steps also include:

[0075] S12. Perform grayscale normalization on the image feature data to construct a standard image matrix, specifically including:

[0076] The gray values ​​of each pixel in the image feature data are extracted, and after pixel statistical processing, the mean and standard deviation of the gray values ​​of all pixels are calculated.

[0077] Based on the mean and standard deviation of the gray values ​​of all pixels, a normalization calculation is performed on the gray values ​​of each pixel in the image feature data to construct a standard image matrix, specifically: In the formula, Fim x,y H represents the standard image matrix. x,y σ represents the gray value of the corresponding pixel, μ represents the mean of the gray values ​​of all pixels, σ represents the standard deviation of the gray values ​​of all pixels, x represents the horizontal coordinate of the corresponding pixel, and y represents the vertical coordinate of the corresponding pixel.

[0078] In this embodiment, a highly structured and coordinate-based image input mechanism is established through standardized processing of the original road surface image and construction of image feature data. This significantly overcomes the problems of blurred crack boundaries and unstable feature recognition caused by uneven illumination, reflection interference, and differences in image grayscale distribution in traditional road detection methods. First, a camera module fixed to the front of the inspection vehicle acquires road surface images using a periodic shutter method. Combined with a GNSS module, geographic coordinate information is recorded synchronously, ensuring a one-to-one correspondence between image data and spatial location. This lays the foundation for subsequent regional comparison, result tracing, and geographic annotation. Then, a two-dimensional pixel coordinate plane is constructed based on pixel resolution and image size, explicitly mapping each pixel in the image to its coordinate position and extracting its grayscale value, thus forming a complete set of image feature data. This structure not only ensures the computability of the image but also enables subsequent grayscale-based difference analysis, boundary judgment, and connectivity analysis. A unified input standard is established. Based on this, statistical analysis of the grayscale value set is performed to further calculate the global mean and standard deviation of the image. This is then used as a basis for normalizing the grayscale of each pixel, constructing a standard image matrix. This effectively suppresses grayscale shifts caused by strong lighting, high-reflectivity areas, or low-brightness shadows. This operation does not introduce weighting coefficients or artificial factors, ensuring the objectivity and adaptability of the normalization results and avoiding grayscale data distortion due to environmental differences. After processing, the generated standard image matrix not only possesses stable grayscale representation but also serves as a unified basis for subsequent crack response calculations and structural analysis. This achieves precise integration between image features and structural modeling, comprehensively improving adaptability and robustness in complex road scenarios. This specific step effectively addresses the key issues raised in the background technology, namely, the difficulty in achieving image feature consistency and the ambiguity of grayscale recognition boundaries, ensuring that subsequent abnormal response judgments are based on stable and reliable data input.

[0079] Example 3

[0080] Please refer to Figures 1 and 3. Specifically, the steps in S2 include:

[0081] S21. Perform directional gray-level difference calculation on the standard image matrix. Perform first-order derivative operations along the x-axis and y-axis respectively to extract the horizontal and vertical gray-level change rates. The specific extraction process includes:

[0082] Centered on each pixel, a 3×3 pixel matrix around the corresponding pixel is selected as the corresponding local pixel region. The gradient response of the corresponding pixel in the horizontal and vertical directions is extracted by the Sobel operator, and the horizontal gray-scale change rate and vertical gray-scale change rate of each pixel are determined respectively.

[0083] The horizontal and vertical grayscale change rates of each pixel are obtained using the following formulas:

[0084] G x =Fim x+1,y -Fim x-1,y G y =Fim x,y+1 -Fim x,y-1 ;

[0085] In the formula, G x With G y These represent the horizontal and vertical grayscale change rates of the corresponding pixels, respectively. x+1,y and Fim x-1,y These represent the grayscale values ​​of the right and left adjacent pixels of the corresponding pixel in the horizontal direction, respectively. x,y+1 and Fim x,y-1 These represent the grayscale values ​​of the adjacent pixels above and below each pixel in the vertical direction, respectively.

[0086] This formula approximates the image using first-order derivative difference, performing gradient operations in two orthogonal directions to capture abrupt changes in grayscale values ​​in local regions, thereby identifying edge or structurally abrupt regions. The background section points out that existing technologies have problems with image feature uniformity, especially under illumination interference and texture overlap, leading to unclear boundaries and blurred features between cracks and normal areas. Therefore, relying solely on static grayscale values ​​for region identification is prone to misjudgment. Introducing this grayscale change rate calculation formula eliminates reliance on the absolute grayscale value of pixels, instead capturing the trend and magnitude of pixel grayscale changes in the spatial neighborhood. When grayscale values ​​change abruptly within a pixel region, the horizontal or vertical change rate increases significantly, forming gradient boundaries; conversely, continuous texture regions show gradual grayscale changes with a change rate close to zero. This directional grayscale difference operation effectively distinguishes structurally broken regions from texture noise regions, effectively solving the problems of "blurred grayscale boundaries" and "misidentification of texture interference" in the background.

[0087] S22. Based on the extracted horizontal and vertical grayscale change rates, and combined with the crack prediction response function, after dimensionless processing, the grayscale gradient change amplitude of each pixel is analyzed to construct an abnormal response matrix, specifically: Afm x,y =|G x -G y |;In the formula, Afm x,y G represents the abnormal response matrix. x With G y These represent the horizontal and vertical grayscale change rates of the corresponding pixels, respectively. The abnormal response matrix includes the abnormal texture response values ​​of the corresponding pixels.

[0088] The constructed anomaly response matrix is ​​used to measure the degree of difference between the horizontal and vertical grayscale change rates of each image pixel, thereby revealing whether it is located in a potential structural anomaly region; G in this formula x With G y These represent the grayscale change rates of the current pixel in the horizontal and vertical directions, respectively, obtained by processing the first derivative of the local pixel region using the Sobel operator. The background technology points out that existing methods struggle to separate the grayscale overlap between cracks and normal textures, especially under conditions of lighting interference or texture reflection, where traditional edge recognition strategies often suffer from misidentification of candidate regions. The core logic of this formula lies in significantly improving the response sensitivity to anomalous structural change regions by solving for the difference in grayscale change amplitude in the two directions. The formula uses absolute value processing to shield directional influences, retaining only the gradient change intensity, thereby establishing a unified-dimensional response index and effectively avoiding the risk of false positives introduced by texture noise. The resulting anomalous response matrix no longer directly relies on pixel grayscale threshold judgment but is constructed based on "structural gradient asymmetry," a crucial step in the transformation of image features into structural features. This lays a clear and engineering-significant data foundation for subsequent crack region connectivity analysis, area screening, and structural index modeling. This processing mechanism strengthens the boundary difference between crack recognition and noise removal, constituting an important entry point for the explicit extraction of anomalous structures in the entire intelligent recognition process.

[0089] Specifically, the S2 steps also include:

[0090] S23. Compare and analyze the abnormal texture response values ​​of each pixel in the abnormal response matrix with preset response thresholds, and use the eight-neighbor connectivity determination method to select several independent candidate connected regions, specifically including:

[0091] If the abnormal texture response value of the corresponding pixel exceeds the preset response threshold, it indicates that the corresponding pixel is a structural texture break and is marked as a candidate pixel; otherwise, it is marked as a normal pixel.

[0092] Based on the labeled candidate pixels, and using the eight-neighbor connectivity method, all consecutive candidate pixels on the plane are aggregated into several independent candidate connected regions, and the abnormal texture response values ​​and coordinate positions of each candidate pixel in the corresponding candidate connected region are recorded.

[0093] The preset response threshold is based on the distribution of abnormal texture response values ​​in crack areas and normal areas in a large number of historical road image samples. The boundary point of abnormal texture response values ​​is determined by statistical analysis, so that it can effectively distinguish typical crack features from natural texture changes, thereby ensuring that the selected candidate pixels have significant structural fracture features.

[0094] The eight-neighbor connectivity determination method is a region aggregation method based on the spatial topological relationship of pixels in an image. Its core purpose is to identify all spatially continuous pixel sets with the same labeling features in an image. This method is used to determine the connectivity of candidate pixel positions in anomaly response matrices, thereby aggregating local pixel perturbations into identifiable candidate connected regions. Specifically, an eight-neighborhood refers to the set of pixel neighborhoods formed by the eight directly adjacent pixels (top, bottom, left, right, and four diagonals) around the center pixel. During the connectivity determination process, all pixels marked as candidate points in the image are scanned one by one to determine whether there are other candidate pixels in their eight-neighborhoods. If so, they are classified into the same connected region. Through recursive traversal, all candidate pixels with continuous connectivity are finally aggregated into an independent connected region and assigned a region number.

[0095] Specifically, the S2 steps also include:

[0096] S24. The area of ​​the corresponding candidate connected region is recorded as the number of all candidate pixels in the selected candidate connected region. The perimeter of the corresponding candidate connected region is recorded as the number of candidate pixels counted on the boundary contour of the corresponding candidate connected region. The principal axis length of the corresponding candidate connected region is recorded as the maximum coordinate distance between two candidate pixels in the corresponding candidate connected region.

[0097] S25. Compare and analyze the area of ​​each candidate connected region to screen out abnormal crack regions and construct crack region feature data, specifically including:

[0098] If the area of ​​the corresponding selected connected region exceeds the preset area threshold, it indicates that the corresponding selected connected region is an abnormal crack region. After the region is numbered, the area, perimeter, principal axis length and coordinate position of each candidate pixel of the corresponding crack connected region are recorded to construct crack region feature data.

[0099] If the area of ​​the corresponding selected connected region does not exceed the preset area threshold, it indicates that the corresponding selected connected region is a normal noise region, and the corresponding noise region is removed.

[0100] The area threshold is preset by statistically analyzing the actual pixel area of ​​crack regions in a large number of labeled road images, and selecting the smallest effective connected area that can significantly distinguish structural cracks from image noise as the boundary standard, thereby ensuring that the selected areas have structural continuity and anomalous significance.

[0101] In this embodiment, a series of actions are performed around a standard image matrix, including directional gray-level change extraction, abnormal response construction, connected region screening, and crack region feature recognition. This solves the problems of blurred crack region boundaries, inaccurate response region extraction, and difficulty in distinguishing cracks from noise mentioned in the background technology. The Sobel operator is used to calculate the horizontal and vertical gray-level change rates of the standard image matrix. The first derivative is used to extract the gray-level gradient response of the image in different directions, accurately capturing the texture fracture trend of each pixel in the spatial structure. This step specifically transforms gray-level changes into mathematical expressions, defining the horizontal and vertical change rates respectively, and constructing a local gradient field through derivative operations, effectively capturing the significant gray-level abrupt changes caused by cracks in road images. Subsequently, the change rates in the two directions are combined to construct an abnormal response matrix, and then the abnormal texture response value of each pixel is analyzed and dimensionless processed to ensure the comparability of response values ​​and to prevent the influence of different image gray-level scales. Based on the comparison analysis of response thresholds, automatic screening is performed. Candidate pixels at the structural fracture location are selected, and an eight-neighborhood connectivity determination method is used to accurately aggregate candidate connected regions with spatial continuity and structural consistency. Unlike traditional methods that rely on fixed thresholds or edge scanning, this method fully utilizes the local response features of gradient changes and spatial connectivity information, avoiding over-segmentation of crack regions or confusion with background textures. Finally, by quantifying and comparing parameters such as the area, perimeter, and principal axis length of connected regions, the true crack region is identified based on the area threshold, and noise regions are accurately eliminated. This series of analyses not only preserves the true structural morphology of the crack but also provides a reliable regional basis for subsequent structural response modeling, significantly improving the accuracy and stability of crack identification. This step clearly addresses the shortcomings of the background technology, namely the lack of crack connectivity analysis and the difficulty in distinguishing between cracks and noise regions. It establishes a complete determination mechanism from gradient identification to structural aggregation and then to region discrimination, providing a structured anomaly extraction method with boundary constraints and morphological parameter support for image recognition.

[0102] Example 4

[0103] Please refer to Figures 1 and 3. Specifically, the steps in S3 include:

[0104] S31. Based on the boundary contours of each crack connected region and combined with the crack region feature data, extract the candidate pixel coordinates of the uppermost, lowermost, leftmost and rightmost edges of the boundary contours of each crack connected region. After calculating the difference in centroid displacement in the vertical and horizontal directions, obtain the vertical centroid displacement difference and the horizontal centroid displacement difference of each crack connected region respectively.

[0105] Vertical and horizontal centroid displacement differences are key parameters used to measure the geometric distribution characteristics of crack connectivity regions in an image. They mainly reflect the degree of centroid offset of the crack region relative to the region boundary in the vertical and horizontal directions. The calculation is based on the positions of the top, bottom, left, and rightmost pixels of the crack region boundary. After statistically analyzing the coordinate centroids of all candidate pixels in the region, the difference is calculated with the central axis formed by these edge points to obtain the vertical and horizontal centroid displacement differences, respectively. These displacement differences can reveal whether the crack region has structural characteristics such as skewness, tilt, or bending, which facilitates the subsequent quantitative processing of structural response strength modeling and crack level assessment. It also provides support for solving the problem of "lack of structural response modeling mechanism" mentioned in the background.

[0106] S32. Correlate the vertical centroid displacement difference with the horizontal centroid displacement difference, the area of ​​the region, and the principal axis length of each crack-connected region. Combine this with the arctangent function to analyze the surface structural response of each crack-connected region and determine the structural response coefficient of each crack-connected region. Specifically: In the formula, Srm i ΔDx represents the structural response coefficient of the corresponding crack-connected region. i and ΔDy i Sr represents the difference in horizontal and vertical centroid displacements of the corresponding crack-connected regions, respectively. i and Lr i These represent the area and principal axis length of the corresponding crack-connected region, respectively. This represents the positive and negative tangent functions, used to characterize the shape and structural offset angle of the corresponding crack-connected region.

[0107] The formula in S32 couples shape offset with scale and structural features to form a structural response coefficient that reflects crack complexity, extension trend, and structural disturbance intensity. Its logical significance lies in effectively characterizing whether a crack is a typical structural failure, rather than random texture disturbance or image noise, by fusing the dual factors of geometric shape and spatial displacement. Traditional image recognition systems proposed in the background technology struggle to quantitatively model crack structural responses, especially in the absence of analysis on displacement and morphological changes within the crack's connected regions, easily leading to misidentification or missed judgment. This formula is designed to address this lack of structural modeling, supplementing the limitations of texture and grayscale features through multi-dimensional fusion analysis, providing key parameter support for subsequent crack strength assessment and automated judgment.

[0108] Specifically, the S3 steps also include:

[0109] S33. Extract the abnormal texture response values ​​of each candidate pixel in each crack connected region, and combine them with the structural response coefficient of the corresponding crack connected region to determine the road surface abnormal intensity index, specifically including:

[0110] Based on the abnormal response matrix constructed by S22, the abnormal texture response values ​​of each candidate pixel in each crack connected region are extracted, and the mean abnormal texture response value of each crack connected region is obtained by combining the statistical mean calculation algorithm.

[0111] By correlating the mean of the abnormal texture response in each crack-connected region with the structural response coefficient, the degree of structural anomaly of the current pavement is analyzed, and the pavement anomaly strength index of the current pavement is obtained, specifically: In the formula, Zyc represents the pavement anomaly strength index, and Srm represents the pavement anomaly strength index i Afm represents the structural response coefficient of the corresponding crack-connected region. i This represents the mean of the anomalous texture response in the corresponding crack-connected region, where n represents the number of crack-connected regions. This indicates the degree of anomaly in the corresponding crack-connected region.

[0112] The above formula is a quantitative index formed by comprehensively evaluating the structural characteristics and texture anomalies of multiple crack connected regions. Its core lies in establishing a non-linear correlation between local structural response and pixel-level abnormal textures to achieve an intensity expression that better reflects the complex structural nature of road cracks. In the formula, Srm... i Afm represents the structural response coefficient of each crack-connected region, reflecting the degree of geometric deviation and structural disturbance in that region. i This represents the mean of the abnormal texture response in the corresponding region, indicating the pixel-level abnormal texture intensity; by introducing 1+log(1+Afm) i The logarithmic form of the formula enables dynamic modulation of local anomaly response values, reducing the dominant effect of abrupt response on overall anomaly evaluation and thus avoiding misjudgment caused by local bright spots or abnormal pixel noise. The summation of all connected crack regions yields the pavement anomaly intensity index, which is the anomaly intensity of the road segment represented by the current detection image. This calculation logic is highly consistent with the problems mentioned in the background technology: traditional methods have difficulty distinguishing between structural deformation and texture interference, and have difficulty generating quantitative results for the overall road condition. This formula, however, achieves step-by-step modeling from the regional level to the overall level, enabling road detection methods to build a bridge between structural modeling and anomaly identification, and to make more robust and targeted judgments on crack distribution.

[0113] In this embodiment, a crack morphology modeling mechanism guided by spatial displacement characteristics is established, focusing on structural response analysis and abnormal intensity index construction of crack regions. This significantly solves the problems in existing technologies where crack severity is difficult to quantify and response degree is inaccurate due to the lack of structural modeling. First, by extracting the boundary contour of the crack connected region, the coordinates of the top, bottom, left, and right candidate pixels are located respectively. Based on this, the difference in centroid displacement of the region in the horizontal and vertical directions is calculated. This processing method can realistically reflect the asymmetry and offset trend of the crack during spatial deformation, providing basic data for structural imbalance identification. The two centroid displacement differences are correlated with the region area and principal axis length, and the arctangent function is introduced to model the response degree, constructing the structural response coefficient of each crack connected region. This coefficient integrates three geometric parameters: spatial offset, region size, and principal axis length, and has good morphological adaptability. It can effectively characterize the structural evolution characteristics of cracks from narrow linear types to extended block types, avoiding the traditional method of relying solely on boundary length or area while ignoring shape. This approach addresses the limitations of static features by extracting abnormal texture response values ​​from candidate pixels within each crack's connected region and obtaining the overall abnormal response feature value of the region through mean statistical analysis. This avoids interference from individual extreme pixel values ​​in the judgment. The mean abnormal response value is combined with the aforementioned structural response coefficient to construct the overall pavement anomaly intensity index for the current pavement cracks. This not only uniformly characterizes the texture changes and geometric structure features of the cracks but also possesses a stable numerical scale, enabling comparison of crack severity across multiple regions. More importantly, the construction of the pavement anomaly intensity index is based on the fusion processing of prior image features and spatial structure, achieving effective cross-layer transmission from pixel-level response to structural-level judgment. This precisely addresses the issues of distorted structural response expression and difficulty in quantifying crack configurations mentioned in the background technology. This step establishes an automated and parameterized anomaly discrimination mechanism through standardized formula modeling and joint structural texture evaluation methods. It provides quantitative support and a unified judgment benchmark for accurate discrimination and subsequent strategy execution in multi-scenario road environments, significantly enhancing the practicality and reliability of road quality detection methods.

[0114] Example 5

[0115] Please refer to Figures 1 and 3. Specifically, the steps in S4 include:

[0116] S41. Compare and analyze the current pavement anomaly strength index with the preset anomaly strength threshold to determine whether there is anomaly in the current pavement structure, and issue the corresponding level of quality inspection results, specifically including:

[0117] If the current pavement abnormality strength index does not exceed the abnormality strength threshold, it indicates that the current pavement structure is normal and a first-level quality monitoring result is issued. At this time, the S1 process is executed to continue collecting the next road surface image.

[0118] If the pavement abnormality strength index exceeds the abnormality strength threshold, it indicates that there is an abnormality in the current pavement structure, and a secondary quality monitoring result is issued. At this time, road maintenance personnel are notified to repair the current road and the latitude and longitude location information of the current road is sent.

[0119] The preset abnormal intensity threshold is based on the statistical results of the pavement abnormal intensity index distribution of a large number of road abnormal samples, combined with the actual crack hazard level standard. By setting a critical value that can effectively distinguish between structural abnormalities and normal fluctuations, the detection method has high sensitivity and false alarm control capability when identifying real crack abnormalities, thereby achieving accurate road quality level determination and subsequent maintenance response.

[0120] In this embodiment, by directly comparing the pavement anomaly intensity index calculated in the previous stage with a preset anomaly intensity threshold, a clear and explicit structural state discrimination system with a two-level response mechanism is realized. This fundamentally solves the problems mentioned in the background technology, such as "lack of closed-loop judgment process," "lack of hierarchical response strategy for anomaly identification," and "severe response delay." Specifically, based on the joint modeling of accurate structural response coefficients and texture anomalies in the early stage, an anomaly intensity index with structural representativeness is generated. Here, by comparing it one by one with the anomaly intensity threshold, it is possible not only to determine whether there is an anomaly in the current pavement, but also to determine the subsequent action logic based on the absolute result of the difference. If the pavement anomaly intensity index does not exceed the anomaly intensity threshold, it is determined that the structure is normal and the pavement state is stable. Then, a first-level quality monitoring result is issued and the next image acquisition process is entered, realizing seamless iterative inspection throughout the entire process, effectively ensuring the continuity of detection and saving resources. Otherwise, it is immediately classified as abnormal. The system identifies abnormal structures and issues a secondary quality monitoring result. This is not merely a data anomaly alert, but a trigger point for system-wide cascading actions. It automatically notifies road maintenance personnel, along with the current GNSS positioning information of the pavement, enabling precise location of the fault area and targeted dispatch response. This mechanism effectively overcomes the false alarm and missed alarm problems caused by ambiguous judgments in traditional methods, and also avoids large-scale, inefficient manual follow-up inspections. S4 links image structure recognition, numerical analysis, and operation and maintenance linkage to form a closed-loop detection process from data collection and analysis to response, significantly improving application efficiency and dynamic adaptability in large-scale road networks. This approach, which combines numerical thresholds with structural judgment, strengthens the automatic discrimination capability and engineering practicality of the detection system, fully addressing the technical shortcomings mentioned in the background, such as "unstable recognition, untimely feedback, and inability to support multiple scenarios," demonstrating innovation and effectiveness in the construction of automated intelligent detection.

[0121] Example 6

[0122] Please refer to Figures 1 and 2. Specifically, a machine recognition-based automated intelligent road quality detection system includes an image preprocessing module, a feature analysis module, a structure evaluation module, and a quality judgment module.

[0123] The image preprocessing module is used to determine image feature data information based on the original road surface image captured by the camera module fixed on the front bracket of the road inspection vehicle, and to construct a standard image matrix after performing grayscale normalization.

[0124] The feature analysis module is used to perform directional gray-level difference calculation on the standard image matrix, and combined with the crack prediction response function, construct an abnormal response matrix. Using the eight-neighbor connectivity determination method, candidate connected regions are selected, the area of ​​each candidate connected region is compared and analyzed, abnormal crack regions are selected, and crack region feature data is constructed.

[0125] The structural assessment module analyzes the surface structural response of each connected region of cracks based on crack region feature data and combined with arctangent function, determines the structural response coefficient of each connected region of cracks, and determines the road surface anomaly intensity index by combining the extracted abnormal texture response values ​​of each candidate pixel in each connected region of cracks.

[0126] The quality assessment module is used to compare and analyze the pavement anomaly strength index with the preset anomaly strength threshold to determine whether there is anomaly in the current pavement structure and issue the corresponding level of quality inspection results.

[0127] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An automated intelligent detection method for road quality based on machine recognition, characterized in that: Includes the following steps: S1. Based on the original road surface images captured by the camera module fixed to the front bracket of the road inspection vehicle, determine the image feature data information, perform grayscale normalization, and construct a standard image matrix; S2. Perform directional grayscale difference calculation on the standard image matrix, and combine it with the crack prediction response function to construct an abnormal response matrix. Use the eight-neighbor connectivity determination method to screen out candidate connected regions, compare and analyze the area of ​​each candidate connected region, screen out abnormal crack regions, and construct crack region feature data; The specific steps of S2 include: S21. Directional gray-level difference calculation is performed on the standard image matrix. First-order derivative operations are performed along the x-axis and y-axis respectively to extract the horizontal and vertical gray-level change rates. Specifically, the extraction process includes: selecting a 3×3 pixel matrix surrounding each pixel as its local pixel region; extracting the gradient response of the corresponding pixel in the horizontal and vertical directions using the Sobel operator; and determining the horizontal and vertical gray-level change rates for each pixel. The horizontal and vertical gray-level change rates for each pixel are obtained using the following formulas: In the formula, and S22. Based on the extracted horizontal and vertical grayscale change rates, and combined with the crack prediction response function, after dimensionless processing, the grayscale gradient change amplitude of each pixel is analyzed to construct an abnormal response matrix, specifically: In the formula, Represents the abnormal response matrix. and These represent the horizontal and vertical grayscale change rates of the corresponding pixels, respectively. The anomaly response matrix includes the anomaly texture response values ​​of the corresponding pixels. S3: Based on the crack region feature data and combined with the arctangent function, analyze the surface structure response degree of each crack connected region, determine the structural response coefficient of each crack connected region, and combine the extracted anomaly texture response values ​​of each candidate pixel in each crack connected region to determine the road surface anomaly intensity index. The specific steps of S3 include: S31: Based on the boundary contour of each crack connected region and combined with the crack region feature data, extract the coordinate positions of the candidate pixels at the top, bottom, left, and right edges of the boundary contour of each crack connected region. After calculating the difference in centroid displacement in the vertical and horizontal directions, obtain the vertical and horizontal centroid displacement differences of each crack connected region. S32: Correlate the vertical and horizontal centroid displacement differences, region area, and principal axis length of each crack connected region. Combine with the arctangent function to analyze the surface structure response degree of each crack connected region and determine the structural response coefficient of each crack connected region. Specifically: In the formula, This represents the structural response coefficient of the corresponding crack-connected region. and These represent the horizontal and vertical centroid displacement differences, respectively, within the connected regions of the corresponding cracks. and These represent the area and principal axis length of the corresponding crack-connected region, respectively. S4 represents the tangent function; compare and analyze the pavement anomaly strength index with the preset anomaly strength threshold to determine whether there is anomaly in the current pavement structure, and issue the corresponding level of quality inspection results.

2. The automated intelligent detection method for road quality based on machine recognition according to claim 1, characterized in that: The specific steps of S1 include: S11, determining image feature data information based on the original road surface image captured by the camera module fixed on the front bracket of the road inspection vehicle, specifically including: during the current road surface image acquisition process, the camera module shutter is triggered according to a preset periodic control signal to capture the current road surface image, and after grayscale processing, the original road surface image is constructed. The latitude and longitude position information of the current road is located in real time through the GNSS receiving unit, and the latitude and longitude position information of the current road and the original road surface image are sent to the local data center for storage; according to the pixel resolution of the original road surface image, the original road surface image is divided into pixels, and combined with the image size, a two-dimensional pixel coordinate plane is established. At the same time, the grayscale value and corresponding coordinate position of each pixel in the two-dimensional pixel coordinate plane are recorded to construct image feature data information.

3. The automated intelligent detection method for road quality based on machine recognition according to claim 2, characterized in that: The specific steps of S1 also include: S12, performing grayscale normalization on the image feature data information to construct a standard image matrix, specifically including: extracting the grayscale value of each pixel in the image feature data information, calculating the mean and standard deviation of the grayscale values ​​of all pixels after pixel statistical processing; and performing normalization calculation on the grayscale values ​​of each pixel in the image feature data information based on the mean and standard deviation of the grayscale values ​​of all pixels to construct a standard image matrix, specifically: In the formula, Represents a standard image matrix. σ represents the gray value of the corresponding pixel, μ represents the mean of the gray values ​​of all pixels, σ represents the standard deviation of the gray values ​​of all pixels, x represents the horizontal coordinate of the corresponding pixel, and y represents the vertical coordinate of the corresponding pixel.

4. The automated intelligent detection method for road quality based on machine recognition according to claim 3, characterized in that: The specific steps of S2 also include: S23, comparing and analyzing the abnormal texture response values ​​of each pixel in the abnormal response matrix with the preset response threshold, and using the eight-neighbor connectivity determination method to select several independent candidate connected regions. Specifically, if the abnormal texture response value of the corresponding pixel exceeds the preset response threshold, it indicates that the corresponding pixel is a structural texture break position and is marked as a candidate pixel; otherwise, it is marked as a normal pixel. Based on the marked candidate pixels, and using the eight-neighbor connectivity determination method, all continuous candidate pixels on the plane are aggregated into several independent candidate connected regions, and the abnormal texture response values ​​and coordinate positions of each candidate pixel in the corresponding candidate connected region are recorded.

5. The automated intelligent detection method for road quality based on machine recognition according to claim 4, characterized in that: S2 further includes the following steps: S24, using the number of all candidate pixels in the selected candidate connected regions as the area of ​​the corresponding candidate connected region, using the number of candidate pixels counted on the boundary contour of the corresponding candidate connected region as the perimeter of the corresponding candidate connected region, and using the maximum coordinate distance between two candidate pixels in the corresponding candidate connected region as the principal axis length of the corresponding candidate connected region; S25, comparing and analyzing the area of ​​each candidate connected region, filtering out abnormal crack regions, and constructing crack region feature data, specifically including: if the area of ​​the corresponding selected connected region exceeds a preset area threshold, it indicates that the corresponding selected connected region is an abnormal crack region, after region numbering, and recording the area, perimeter, principal axis length, and coordinate positions of each candidate pixel of the corresponding crack connected region, constructing crack region feature data; if the area of ​​the corresponding selected connected region does not exceed the preset area threshold, it indicates that the corresponding selected connected region is a normal noise region, and the corresponding noise region is removed.

6. The automated intelligent detection method for road quality based on machine recognition according to claim 5, characterized in that: S3 further includes the following steps: S33, extracting the abnormal texture response values ​​of each candidate pixel in each crack connected region, and determining the pavement anomaly intensity index by combining the structural response coefficient of the corresponding crack connected region. Specifically, this includes: extracting the abnormal texture response values ​​of each candidate pixel in each crack connected region based on the anomaly response matrix constructed in S22, obtaining the mean abnormal texture response value of each crack connected region by combining a statistical averaging algorithm; correlating the mean abnormal texture response value of each crack connected region with the structural response coefficient to analyze the degree of structural anomaly of the current pavement and obtain the pavement anomaly intensity index of the current pavement. In the formula, Zyc represents the pavement anomaly strength index. This represents the structural response coefficient of the corresponding crack-connected region. represents the mean of the abnormal texture response of the corresponding crack connected region, and n represents the number of crack connected regions.

7. The automated intelligent detection method for road quality based on machine recognition according to claim 6, characterized in that: The specific steps of S4 include: S41, comparing and analyzing the current pavement surface anomaly strength index with the preset anomaly strength threshold to determine whether there is an anomaly in the current pavement structure, and issuing the corresponding level of quality inspection results. Specifically, if the current pavement surface anomaly strength index does not exceed the anomaly strength threshold, it indicates that the current pavement structure is normal, and a level one quality monitoring result is issued. At this time, the S1 process is executed to continue collecting the next road surface image; if the current pavement surface anomaly strength index exceeds the anomaly strength threshold, it indicates that there is an anomaly in the current pavement structure, and a level two quality monitoring result is issued. At this time, road maintenance personnel are notified to repair the current road, and the latitude and longitude location information of the current road is sent.

8. A machine-recognition-based automated intelligent road quality detection system, used to implement the machine-recognition-based automated intelligent road quality detection method according to any one of claims 1 to 7, characterized in that: It includes an image preprocessing module, a feature analysis module, a structure evaluation module, and a quality judgment module. The image preprocessing module is used to determine image feature data information based on the original road surface image captured by the camera module fixed on the front bracket of the road inspection vehicle, and after performing grayscale normalization, construct a standard image matrix. The feature analysis module is used to perform directional gray-level difference calculation on the standard image matrix, and combined with the crack prediction response function, construct an abnormal response matrix. Using the eight-neighbor connectivity determination method, candidate connected regions are selected, the area of ​​each candidate connected region is compared and analyzed, abnormal crack regions are selected, and crack region feature data is constructed. The structural assessment module analyzes the surface structural response of each connected region of cracks based on crack region feature data and combined with arctangent function, determines the structural response coefficient of each connected region of cracks, and determines the road surface anomaly intensity index by combining the extracted abnormal texture response values ​​of each candidate pixel in each connected region of cracks. The quality assessment module is used to compare and analyze the pavement anomaly strength index with the preset anomaly strength threshold to determine whether there is anomaly in the current pavement structure and issue the corresponding level of quality inspection results.

Images