Rural highway pavement disease intelligent identification and positioning system

Abstract

The invention relates to the field of image processing, and particularly discloses a rural highway pavement disease intelligent identification and positioning system comprising an image standardization module used for obtaining a standardized grayscale image; the wavelet decomposition module is used for obtaining a low-frequency sub-band and a plurality of high-frequency sub-bands; the high-frequency processing module is used for carrying out soft threshold processing on the high-frequency coefficient and retaining high-variance region features; the inverse transformation module is used for reconstructing the de-noised image; the edge enhancement module is used for highlighting crack and pit slot target contour information; the binarization module is used for segmenting a foreground disease candidate region; and the edge filling module is used for separating the disease from the background. According to the method, the core contradiction between background impurity removal and disease feature retention in rural highway tiny disease recognition is effectively solved, a traditional denoising algorithm either smooths tiny disease features or cannot thoroughly remove background impurities, and the system achieves the balance of the background impurity removal and the disease feature retention through cooperation of multiple modules.

Description

Technical Field

[0001] This invention relates to the field of image processing, specifically to an intelligent identification and location system for rural road surface defects. Background Technology

[0002] In rural road defects, features such as hairline cracks are extremely subtle. Traditional denoising algorithms, such as Gaussian filtering, smooth out these subtle defects simultaneously. However, adjusting the algorithm to preserve the defects cannot completely remove background impurities such as fallen leaves. For example, when denoising pothole images on gravel roads, algorithms easily misjudge the gravel texture at the pothole edges as noise, leading to distortion of the pothole contours. This problem seriously affects the accuracy of rural road surface defect identification and hinders subsequent defect localization and maintenance. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide an intelligent identification and location system for rural road surface defects.

[0004] This invention provides an intelligent identification and location system for rural road surface defects, comprising: The image normalization module is used to convert the acquired color image into a grayscale image, unify the image size and brightness, and obtain a standardized grayscale image; The wavelet decomposition module is used to perform two-dimensional discrete wavelet decomposition on the standardized grayscale image to obtain low-frequency subbands and multiple high-frequency subbands. The high-frequency processing module is used to calculate the local variance for each high-frequency sub-band, set an adaptive threshold, perform soft thresholding on the high-frequency coefficients, and retain the characteristics of high-variance regions. The inverse transform module is used to perform wavelet inverse transform on the processed high-frequency subband and low-frequency subband to reconstruct the denoised image; The edge enhancement module is used to enhance the contrast of disease edges in denoised images and highlight the outline information of cracks and pits. The binarization module is used to calculate the global threshold of the enhanced image, binarize the enhanced image, and segment out the candidate regions for foreground diseases. The edge filling module is used to connect broken edges in a binarized image and fill holes, separating the lesion from the background.

[0005] Furthermore, the process of converting the acquired color image into a grayscale image and standardizing the image size and brightness to obtain a standardized grayscale image involves: acquiring a color image from a vertical top-down view using a high-definition camera on a mobile platform; converting the acquired color image into a grayscale image that retains brightness information; standardizing the grayscale image size; adjusting the brightness and contrast; and eliminating the effects of uneven lighting to obtain a standardized grayscale image.

[0006] Furthermore, the two-dimensional discrete wavelet decomposition of the standardized grayscale image specifically involves: selecting wavelet basis functions that preserve edge features, setting the number of decomposition layers, iteratively performing high-pass and low-pass filtering on the image, and downsampling to generate sub-band images for each layer.

[0007] Furthermore, for each high-frequency sub-band, the local variance is calculated and an adaptive threshold is set. Specifically, for each high-frequency sub-band, the detailed information of all directions and decomposition layers is traversed, the local variance is calculated, and an adaptive threshold that is inversely proportional to the local variance is set. Regions with large local variances have higher thresholds.

[0008] Furthermore, the process of performing inverse wavelet transform on the processed high-frequency subband and the low-frequency subband specifically involves recombining the processed high-frequency subband and the unprocessed low-frequency subband according to the original decomposition structure to form the complete coefficient matrix required for the inverse wavelet transform. Using the same wavelet basis functions and decomposition levels as the wavelet decomposition, the inverse transform algorithm processes and iterates layer by layer in the reverse order of decomposition. The layer-by-layer processing involves first upsampling the current layer subband, then performing corresponding filtering, and finally linearly combining to recover the image data of the previous layer.

[0009] Furthermore, the method of enhancing the contrast of disease edges in the denoised image and highlighting the contour information of cracks and pits involves applying the Laplacian Gaussian operator to the multi-scale adaptively denoised image for edge enhancement. The operator first Gaussian smooths the image and then calculates the second derivative. The standard deviation of the Gaussian kernel is selected according to the disease feature scale. During the calculation, the denoised image is first Gaussian convolved, and then the Laplacian operator is applied to obtain the Laplacian Gaussian response map.

[0010] Furthermore, the calculation of the global threshold involves binarizing the enhanced image and segmenting the foreground disease candidate region. Specifically, the image enhanced by the Gaussian Laplacian operator is first subjected to grayscale consistency calibration. Then, the gray levels of the enhanced image are traversed using the maximum inter-class variance method. The foreground and background are divided according to the candidate gray levels. The proportion and average gray value of the foreground and background are statistically analyzed, and the inter-class variance is calculated. The gray level with the largest inter-class variance is selected as the global optimal threshold. Finally, the image is binarized with this threshold, and the foreground is marked as a disease candidate region and the background is marked as a non-disease region, thus achieving preliminary separation of the foreground and background.

[0011] Furthermore, the process of connecting the fractured edges and filling the holes to separate the disease from the background involves first dilating and then eroding the binarized image. The dilution operation expands the foreground region to connect the fine fractured edges of the disease, while the erosion operation shrinks the foreground region to restore the over-expanded contour and fill the internal holes, ultimately resulting in a binarized image with a continuous contour of the diseased area, no holes, and a clear boundary with the background.

[0012] The beneficial effects of this invention are: This invention effectively solves the core contradiction between removing background impurities and preserving disease features in the identification of minor defects on rural roads. Traditional denoising algorithms either smooth out minor disease features or fail to completely remove background impurities, while this system achieves a balance between the two through multi-module collaboration.

[0013] This invention first eliminates interference from uneven lighting and size differences through image standardization, laying a unified foundation for subsequent processing. Then, it uses wavelet decomposition to split image frequency information, employing adaptive thresholding based on local variance for high-frequency subbands containing disease details. This enhances denoising in low-variance noise regions while preserving key details in high-variance disease feature regions, preventing the false removal of disease features. Subsequently, it reconstructs a high-quality denoised image through inverse transform, further enhancing disease edge contrast using the Laplacian of Gaussian operator, making subtle disease outlines clearer. Next, it achieves precise segmentation of disease candidate regions using an optimal global threshold. Finally, morphological operations connect fracture edges and fill holes, completely separating the disease from the background. This entire process ensures accurate capture of subtle disease features such as hairline cracks and pothole edges while effectively removing background impurities such as fallen leaves and gravel textures, avoiding distortion of disease outlines, ultimately achieving accurate identification and location of pavement diseases on rural roads. Attached Figure Description

[0014] Figure 1 A block diagram of the intelligent identification and positioning system for rural road surface defects of the present invention. Detailed Implementation

[0015] In its specific implementation, this application discloses an intelligent identification and location system for rural road surface defects, such as... Figure 1 The intelligent identification and location system for rural road surface defects includes: The image normalization module is used to convert the acquired color image into a grayscale image, unify the image size and brightness, and obtain a standardized grayscale image; The wavelet decomposition module is used to perform two-dimensional discrete wavelet decomposition on the standardized grayscale image to obtain low-frequency subbands and multiple high-frequency subbands. The high-frequency processing module is used to calculate the local variance for each high-frequency sub-band, set an adaptive threshold, perform soft thresholding on the high-frequency coefficients, and retain the characteristics of high-variance regions. The inverse transform module is used to perform wavelet inverse transform on the processed high-frequency subband and low-frequency subband to reconstruct the denoised image; the edge enhancement module is used to enhance the contrast of the disease edges in the denoised image and highlight the target contour information of cracks and pits. The binarization module is used to calculate the global threshold of the enhanced image, binarize the enhanced image, and segment out the candidate regions for foreground diseases. The edge filling module is used to connect broken edges in a binarized image and fill holes, separating the lesion from the background.

[0016] The process involves converting the acquired color images into grayscale images and standardizing their size and brightness to obtain standardized grayscale images. Specifically, this is achieved by first converting RGB color images acquired from a high-definition CCD camera mounted on a mobile platform such as a vehicle or drone, viewed from a vertical top-down angle, to grayscale. A weighted average method is used to retain brightness information in the image, converting the color images into grayscale images. Next, the resulting grayscale images are standardized in size, adjusting all grayscale images to a uniform pixel resolution to ensure consistency in subsequent processing. Finally, a histogram equalization algorithm is applied to adjust the brightness of the standardized grayscale images, optimizing image contrast by redistributing the grayscale levels of image pixels and eliminating the impact of uneven lighting in the acquisition environment, ultimately yielding standardized grayscale images.

[0017] The process involves performing two-dimensional discrete wavelet decomposition on a standardized grayscale image to obtain a low-frequency sub-band and multiple high-frequency sub-bands. Specifically, this involves performing two-dimensional discrete wavelet decomposition on the standardized grayscale image to obtain multi-scale frequency information. First, a suitable wavelet basis function for edge feature preservation is selected, such as the Daubechies wavelet basis (db4), which has good time-frequency localization characteristics and can effectively distinguish between disease edges and random noise. The number of decomposition layers is set to three to balance computational efficiency and feature preservation capability. The decomposition process uses the Mallat fast decomposition algorithm, iteratively high-pass and low-pass filtering the image and performing a 2x downsampling to generate sub-band images for each layer. Finally, one low-frequency sub-band (LL) and 3L high-frequency sub-bands (HL, LH, HH) are obtained, where L is the number of decomposition layers. The low-frequency sub-band LL reflects the overall contour and approximate information of the image, while the high-frequency sub-bands HL, LH, and HH correspond to detailed information in the horizontal, vertical, and diagonal directions, respectively, including disease features such as crack edges and texture abrupt changes. These multi-scale subbands will serve as the basis for subsequent adaptive thresholding to achieve a balance between noise suppression and feature preservation.

[0018] For each high-frequency sub-band, the local variance is calculated, an adaptive threshold is set, and soft thresholding is applied to the high-frequency coefficients to preserve the features of high-variance regions. Specifically, this involves first traversing each high-frequency sub-band, including all decomposition layers in the horizontal (HL), vertical (LH), and diagonal (HH) directions, to ensure that detailed information at all directions and scales is processed. For each sub-band, a sliding window is used to calculate the local variance.

[0019] For example, the window size is set to 3×3 pixels to capture local detail changes such as disease edges. The local variance is calculated by taking the mean and squared mean of the wavelet coefficients within the window, and then applying the formula Var=mean(X 2 )-(mean(X))2 The local variance is obtained. In the formula, Var represents the local variance within the "sliding window" in the high-frequency sub-band, mean(X) represents the arithmetic mean of all "original high-frequency wavelet coefficients" within the sliding window, mean(X 2 ) represents the arithmetic mean of the squares of all "original high-frequency wavelet coefficients" within the sliding window, and X represents a single "original high-frequency wavelet coefficient" within the sliding window in the high-frequency sub-band (HL / LH / HH directions).

[0020] Based on the Bayesian estimation principle, the threshold is set to be inversely proportional to the local variance. Specifically, in areas with a relatively large local variance (which may be the edges of diseases or textures), the threshold is increased to retain features; in areas with a relatively small local variance (which may be uniform noise), the threshold is decreased to enhance the denoising effect.

[0021] Exemplarily, the threshold calculation formula is T = α / sqrt(Var), where α is a global adjustment parameter that can be dynamically optimized according to the noise level, T represents the "adaptive denoising threshold" for the current sliding window, sqrt(Var) represents the arithmetic square root of the local variance Var (i.e., the local standard deviation), and Var is the same as Var in the "local variance calculation formula" (the local variance within the sliding window). When performing soft threshold processing on the high-frequency coefficients, the formula is used: if |x| ≥ T, then x' = sign(x) × (|x| - T); if |x| < T, then x' = 0. In the formula, x' represents the "output high-frequency wavelet coefficient" after soft threshold processing, x represents the "original high-frequency wavelet coefficient" before processing (the same as X in the local variance formula), and T is the same as T in the "adaptive threshold formula" (the adaptive denoising threshold for the current window). This soft threshold function can smoothly retain significant features and suppress noise. Through the above processing, background impurities are effectively removed, and the disease features in the high-variance areas are ensured to be retained, laying a foundation for subsequent disease identification and localization.

[0022] "Calculating the local variance for each high-frequency sub-band, setting the adaptive threshold, performing soft threshold processing on the high-frequency coefficients, and retaining the features in the high-variance areas" constructs the adaptive threshold only relying on the single feature of the local variance, and there are two limitations: one is that it is impossible to distinguish between "the edges of diseases with high variance" and "background impurities with high variance (such as gravel textures)", which is likely to result in the mis-retention of impurities; the other is that the "rigid reduction" of the soft threshold function to the coefficients will cause the loss of details at the edges of diseases.

[0023] Therefore, this application further proposes a high-frequency subband optimization method based on multi-feature fusion and differentiable shrinkage. By fusing the dual features of "local gradient consistency" and "non-local similarity" and combining a differentiable shrinkage function, it can accurately identify the differences between disease characteristics and impurities, and also achieve smooth adjustment of coefficients. First, it traverses the high-frequency subbands of all decomposition layers in the horizontal (HL), vertical (LH), and diagonal (HH) directions, and inputs the original high-frequency wavelet coefficient matrix of each subband. ( Using a sliding window (5×5 pixels, balancing detail capture and anti-interference) to sequentially extract multiple features within each window (for sub-band size), a dynamic threshold is constructed and differentiable shrinkage processing is performed, ultimately outputting the optimized high-frequency sub-band coefficient matrix. .

[0024] include: For sliding windows (center coordinates) The set of coefficients within ) Calculate three core features: Local variance ( ): Describes the dispersion of the coefficient distribution; both diseases and impurities can exhibit high variance. ; in The total number of pixels in the window. This represents a single original high-frequency wavelet coefficient within the window.

[0025] Local gradient consistency ( ): To distinguish between "continuous disease edges" and "chaotic impurity textures," the gradient direction of disease edges is more consistent. First, calculate the gradient magnitude of each coefficient within the window. Then calculate the gradient direction consistency: ; in The maximum gradient magnitude within the window. This is the normalized mean of the gradient magnitudes within the window. A larger value indicates a more uniform gradient direction (more likely to be a disease).

[0026] Nonlocal similarity ( ): Utilizing image redundancy, the diseased area contains more similar blocks, while impurities are mostly isolated distributions. Searching for and in the current high-frequency subband A set of similar blocks of the same size Calculate the similarity weights: ; in For Euclidean distance, This is a similarity decay factor (with a value of 0.1 to 0.3, controlling the sensitivity of the similarity weight). The larger the value, the more likely the area is to have diseases with redundant characteristics.

[0027] A threshold is constructed based on the weighted fusion result of the above three features to highlight the disease characteristics of "high variance + high gradient consistency + high nonlocal similarity" and suppress impurities and noise. ; in This is a global scale parameter (dynamically adjusted according to the noise level, with a value of 10~30). The feature enhancement coefficient (values ​​2 to 5, amplifying the effects of gradient consistency and nonlocal similarity) is used. To avoid the minimum value where the denominator is zero, For window The corresponding adaptive threshold.

[0028] By using a differentiable function instead of the traditional soft threshold, the loss of detail caused by abrupt changes in coefficients is avoided, achieving a smooth balance between noise suppression and feature preservation. ; in These are the processed high-frequency wavelet coefficients. For hyperbolic tangent function (in) When the time approaches 1, the coefficients are preserved in their original form; The threshold value approaches 0 to suppress noise; a smooth transition occurs near the threshold to avoid edge distortion.

[0029] Output: After traversing all sliding windows of all high-frequency subbands, the optimized high-frequency subband coefficient matrices for each direction and level are obtained. This matrix removes impurities with low feature fusion values, such as fallen leaves and random noise, while preserving subtle disease features with high feature fusion values, such as hairline cracks and pit edges, providing higher quality frequency domain data for subsequent wavelet inverse transform and disease segmentation.

[0030] The process involves performing an inverse wavelet transform on the processed high-frequency and low-frequency subbands to reconstruct the denoised image. Specifically, this involves recombining all the high-frequency subbands after soft thresholding with the unthresholded low-frequency subbands according to the original decomposition structure, forming the complete coefficient matrix required for the inverse wavelet transform. The wavelet basis function (i.e., Daubechies wavelet basis db4) and the number of decomposition levels (3 levels) are used, identical to those in the decomposition stage, to ensure the accuracy of the reconstruction.

[0031] The reconstruction process employs the Mallat fast inverse transform algorithm, proceeding layer by layer in the reverse order of decomposition. Specifically, the low-frequency and high-frequency sub-bands of the current layer are first upsampled by a factor of 2, followed by corresponding low-pass and high-pass filtering. Finally, the filtering results are linearly combined to reconstruct the image data of the previous layer. This process iterates layer by layer until an image with the same size as the input is completely reconstructed from the lowest-resolution low-frequency sub-band and all high-frequency sub-bands. After reconstruction, the output image is normalized in grayscale (e.g., mapped to 0-255) to eliminate numerical offsets caused by floating-point operations and ensure visual consistency.

[0032] Through the above steps, a denoised image is finally obtained that effectively suppresses background impurities while perfectly preserving subtle disease features such as hairline cracks and pit edges, providing high-quality basic data for subsequent disease feature enhancement and identification.

[0033] The method involves enhancing the contrast of disease edges in the denoised image to highlight the contour information of cracks and pits. Specifically, this is achieved by using the Laplacian of Gaussian operator to enhance the edges of the image after multi-scale adaptive denoising. This operator effectively suppresses residual noise while highlighting edge areas with drastic grayscale changes by first Gaussian smoothing the image and then calculating its second derivative.

[0034] For example, firstly, an appropriate Gaussian kernel standard deviation σ is selected based on the scale of the disease characteristics. For fine features such as hairline cracks, a smaller σ value (e.g., 0.5) is chosen to preserve finer edge details; for larger-scale diseases such as pits, a larger σ value (e.g., 1.0) is chosen to enhance their contours. During the calculation process, the denoised image is first subjected to Gaussian convolution to smooth the image and reduce high-frequency noise; then, the Laplacian operator is applied to obtain a Gaussian Laplacian response map, where the edge regions exhibit alternating positive and negative response values.

[0035] To highlight the outline of the disease, zero-cross detection can be performed on the response map to locate the edge position, or the response map can be directly thresholded to generate a binary edge map.

[0036] Through the above treatment, the edge contrast of defects such as cracks and pits is significantly improved, and the outlines are clearer, laying a good foundation for subsequent defect area segmentation.

[0037] The calculation of the global threshold involves binarizing the enhanced image and segmenting the foreground disease candidate region. Specifically, this involves first performing grayscale consistency calibration on the disease edge image enhanced by the Gaussian Laplacian operator to ensure that the image grayscale value remains within the 0-255 range, consistent with the grayscale range of the previous image grayscale and standardization processes, thus avoiding the impact of grayscale fluctuations on the accuracy of the threshold calculation. Then, the maximum inter-class variance method is initiated, traversing all possible grayscale levels (0 to 255) of the enhanced image. For each candidate grayscale level, the image pixels are divided into two categories: the assumed foreground (potential disease region) and the background.

[0038] For each candidate gray level, the proportion of foreground pixels in the entire image and the proportion of background pixels are calculated separately. At the same time, the average gray value of the foreground pixels and the average gray value of the background pixels are calculated. Then, the inter-class variance is calculated according to the formula (inter-class variance = foreground proportion × (foreground average gray value - global average gray value)). 2 +Background percentage × (Average background grayscale - Average global grayscale) 2 The inter-class variance value corresponding to each candidate gray level is calculated one by one. The candidate gray level with the largest inter-class variance is selected and determined as the global optimal threshold. This threshold can highlight the gray level difference between the diseased area and the background to the greatest extent.

[0039] Finally, the enhanced image is binarized using this global threshold. Specifically, pixels with gray values ​​greater than or equal to the threshold are marked as foreground (i.e., disease candidate regions, clearly showing the enhanced disease edges and core areas), while pixels with gray values ​​less than the threshold are marked as background. This achieves preliminary separation between the foreground disease candidate regions and the background, laying the foundation for subsequent morphological closing operations to connect broken edges and fill holes, ensuring the consistency of the segmentation results with the overall disease identification process. The process involves connecting the fractured edges and filling the holes to separate the disease from the background. Specifically, in implementation, for the image after Otsu's inter-class variance binarization, a morphological closing operation of "dilation followed by erosion" is used to address potential issues such as fractured disease edges and small holes within candidate regions, ensuring complete separation of the disease from the background. The selected structural element is a 3×3 rectangle, a size suitable for the initial disease region segmentation, applicable to both fine diseases like hairline cracks and the contour features of pitted diseases.

[0040] For example, firstly, a dilation operation is performed, traversing the binarized image with a 3×3 rectangular structuring element. When the center of the structuring element covers the foreground pixel, all background pixels within the range of the structuring element are updated to foreground pixels. By expanding the foreground region, the fracture edges of minor defects such as hairline cracks are connected, and the tiny gaps between regions are eliminated.

[0041] Then, the erosion operation is performed, using the same 3×3 rectangular structuring element to traverse the dilated image again. When there are background pixels within the structuring element range, the foreground pixel in the center is updated to a background pixel, thereby shrinking the foreground area and restoring the outline of the non-disease area that was over-expanded during the dilation process. This ensures that the size of the disease area is consistent with the original features, while filling the small holes inside the disease candidate area, such as the tiny blanks caused by segmentation in the pit area.

[0042] After this processing, a binarized image is finally obtained with a continuous outline of the foreground disease area, no internal holes, and a clear boundary with the background area. This provides an accurate and complete image data foundation for subsequent extraction of geometric features of the disease area (such as area, perimeter, aspect ratio, etc.) and disease type identification.

[0043] This invention effectively solves the core contradiction between "background impurity removal" and "disease feature preservation" in the identification of minor road defects (such as hairline cracks). Traditional denoising algorithms either smooth out minor disease features or fail to completely remove background impurities, while this system achieves a balance between the two through multi-module collaboration. First, image standardization eliminates interference from uneven lighting and size differences, laying a unified foundation for subsequent processing. Then, wavelet decomposition is used to split the image frequency information. For high-frequency subbands containing disease details, adaptive thresholding based on local variance is applied to enhance denoising in low-variance noise regions and preserve key details in high-variance disease feature regions, avoiding the false removal of disease features. Subsequently, inverse transform is used to reconstruct a high-quality denoised image, and the Laplacian of Gaussian operator is combined to further enhance the contrast of disease edges, making the outlines of minor diseases clearer. Next, the optimal global threshold is used to achieve accurate segmentation of disease candidate regions. Finally, morphological operations are used to connect fracture edges and fill holes, completely separating the disease from the background. The entire process ensures accurate capture of subtle defects such as hairline cracks and pothole edges, while effectively removing background impurities such as fallen leaves and gravel textures, avoiding distortion of the defect outline, and ultimately achieving accurate identification and location of rural road surface defects, overcoming the limitations of traditional treatment methods.

[0044] It is evident that the system of the present invention can be implemented by a computer program, and the computer program for implementing the system of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to the processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer program causes the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer program can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0045] Therefore, it can be understood that this invention discloses an electronic device, comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed, enable the at least one processor to perform the system described above.

[0046] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0047] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A smart identification and location system for rural road surface defects, characterized in that, include, The image normalization module is used to convert the acquired color image into a grayscale image, unify the image size and brightness, and obtain a standardized grayscale image; The wavelet decomposition module is used to perform two-dimensional discrete wavelet decomposition on the standardized grayscale image to obtain low-frequency subbands and multiple high-frequency subbands. The high-frequency processing module is used to calculate the local variance for each high-frequency sub-band, set an adaptive threshold, perform soft thresholding on the high-frequency coefficients, and retain the characteristics of high-variance regions. The inverse transform module is used to perform wavelet inverse transform on the processed high-frequency subband and low-frequency subband to reconstruct the denoised image; The edge enhancement module is used to enhance the contrast of disease edges in denoised images and highlight the outline information of cracks and pits. The binarization module is used to calculate the global threshold of the enhanced image, binarize the enhanced image, and segment out the candidate regions for foreground diseases. The edge filling module is used to connect broken edges in a binarized image and fill holes, separating the lesion from the background.

2. The intelligent identification and positioning system for rural road surface defects according to claim 1, characterized in that, The process of converting the acquired color image into a grayscale image and standardizing the image size and brightness to obtain a standardized grayscale image involves: acquiring a color image from a vertical top-down view using a high-definition camera on a mobile platform; converting the acquired color image into a grayscale image that retains brightness information; standardizing the grayscale image size; adjusting the brightness and contrast; and eliminating the effects of uneven lighting to obtain a standardized grayscale image.

3. The intelligent identification and positioning system for rural road surface defects according to claim 1, characterized in that, The two-dimensional discrete wavelet decomposition of the standardized grayscale image is specifically performed by: selecting wavelet basis functions that preserve edge features, setting the number of decomposition layers, iteratively performing high-pass and low-pass filtering on the image, and downsampling to generate sub-band images for each layer.

4. The intelligent identification and location system for rural road surface defects according to claim 1, characterized in that, For each high-frequency sub-band, the local variance is calculated and an adaptive threshold is set. Specifically, for each high-frequency sub-band, the detailed information of all directions and decomposition layers is traversed, the local variance is calculated, and an adaptive threshold that is inversely proportional to the local variance is set. Regions with large local variances have higher thresholds.

5. The intelligent identification and positioning system for rural road surface defects according to claim 1, characterized in that, The process of performing inverse wavelet transform on the processed high-frequency subband and low-frequency subband involves recombining the processed high-frequency subband and the unprocessed low-frequency subband according to the original decomposition structure to form the complete coefficient matrix required for the inverse wavelet transform. The same wavelet basis functions and decomposition levels as the wavelet decomposition are used. The inverse transform algorithm processes and iterates layer by layer in the reverse order of decomposition. The layer-by-layer processing first performs upsampling on the current layer subband, then performs corresponding filtering, and finally linearly combines to recover the image data of the previous layer.

6. The intelligent identification and positioning system for rural road surface defects according to claim 1, characterized in that, The method of enhancing the contrast of disease edges in the denoised image and highlighting the contour information of cracks and pits involves using the Laplacian Gaussian operator to enhance the edges of the multi-scale adaptively denoised image. The operator first Gaussian smooths the image and then calculates the second derivative. The standard deviation of the Gaussian kernel is selected according to the disease feature scale. During the calculation, the denoised image is first Gaussian convolved and then the Laplacian operator is applied to obtain the Laplacian Gaussian response map.

7. The intelligent identification and positioning system for rural road surface defects according to claim 1, characterized in that, The calculation of the global threshold involves binarizing the enhanced image and segmenting the foreground disease candidate region. Specifically, the enhanced image is first calibrated for grayscale consistency. Then, the grayscale levels of the enhanced image are traversed using the maximum inter-class variance method. The foreground and background are divided according to the candidate grayscale levels. The proportion and average grayscale value of the foreground and background are statistically analyzed, and the inter-class variance is calculated. The grayscale level with the largest inter-class variance is selected as the global optimal threshold. Finally, the image is binarized with this threshold, and the foreground is marked as a disease candidate region and the background is marked as a non-disease region, thus achieving preliminary separation of the foreground and background.

8. The intelligent identification and positioning system for rural road surface defects according to claim 1, characterized in that, The process of connecting the broken edges and filling the holes to separate the disease from the background involves first dilating and then eroding the binarized image. The dilution operation expands the foreground region to connect the fine disease broken edges, while the erosion operation shrinks the foreground region to restore the over-expanded contour and fill the internal holes, ultimately resulting in a binarized image with a continuous contour of the diseased area, no holes, and a clear boundary with the background.

Images