Pavement disease analysis method based on disease feature enhancement and image classification model

Abstract

This invention discloses a pavement defect analysis method based on defect feature enhancement and image classification models, belonging to the field of pavement defect analysis. The method includes: acquiring an initial dataset of pavement images; the initial dataset including defective pavement images and pavement defect information; constructing a pavement defect image classification model; training the pavement defect image classification model using the initial dataset to obtain a trained pavement defect image classification model; acquiring a pavement defect image to be tested; and using the trained pavement defect image classification model to obtain the defect classification result of the pavement defect image to be tested. This invention solves the problem of high image quality requirements in existing pavement defect identification methods.

Description

Technical Field

[0001] This invention belongs to the field of pavement distress analysis, and in particular relates to a pavement distress analysis method based on distress feature enhancement and image classification model. Background Technology

[0002] With the rapid development of my country's economy, the scale and complexity of the road transportation network are also constantly increasing. The quality and safety of roads are of great significance for ensuring smooth traffic flow, reducing traffic accidents, and improving people's travel efficiency. However, due to various natural and human factors, road surfaces develop various defects, such as cracks, potholes, and loose pavement, which seriously affect the service life of roads and traffic safety.

[0003] Road defects, such as cracks, potholes, and loose pavement, not only affect the service life of roads but also seriously impact traffic safety. For example, if road cracks are not repaired in time, they may widen and eventually form potholes, which will seriously hinder vehicle movement and may even cause traffic accidents. Furthermore, loose pavement may cause vehicles to lose control, thus increasing the risk of traffic accidents.

[0004] Therefore, timely and accurate detection and repair of road defects are crucial. Firstly, timely repair can prevent further damage, thus extending the road's lifespan. For example, repairing cracks in their early stages can prevent them from widening, avoiding larger-scale road damage. This will help increase the road's lifespan, thereby reducing the construction costs of new roads.

[0005] Traditional road defect detection methods primarily rely on manual inspections, which are inefficient, costly, and ill-suited for large-scale road networks. In recent years, with the development of computer vision and deep learning technologies, the automatic detection and identification of road defects using these techniques has become a research hotspot. However, these methods typically require large amounts of labeled data, which is both time-consuming and labor-intensive to acquire. Furthermore, due to the diversity and complexity of road surface images, improving the generalization ability of the models to effectively identify road defects in various environments and conditions remains a significant challenge. Summary of the Invention

[0006] To address the aforementioned shortcomings in existing technologies, this invention provides a pavement disease analysis method based on disease feature enhancement and image classification models, which solves the problem that existing pavement disease identification methods have high image requirements.

[0007] To achieve the aforementioned objectives, the technical solution adopted by this invention is: a pavement distress analysis method based on distress feature enhancement and image classification models, comprising:

[0008] Obtain an initial dataset of road surface images; the initial dataset of road surface images includes images of damaged road surfaces and information on road surface defects.

[0009] A road surface defect image classification model is constructed; the road surface defect image classification model is trained using an initial dataset of road surface images to obtain a trained road surface defect image classification model; the road surface defect image classification model includes a defect feature enhancement module, a feature map cleaning module, a feature map clustering module, a feature extraction module, several texture enhancement modules, and a classifier;

[0010] The defect feature enhancement module is used to enhance the defect features of defective pavement images in the initial pavement image dataset to obtain an enhanced set of defective pavement images.

[0011] The feature map cleaning module is used to remove interfering features from the enhanced set of damaged road surface images to obtain a cleaned set of damaged road surface images.

[0012] The feature map clustering module is used to divide the cleaned image set of damaged road surfaces into easily separable image subsets and difficult-to-separate image subsets; the difficult-to-separate image subsets include several difficult-to-separate image clusters.

[0013] The feature extraction module is used to extract features from each image in the easily separable image subset to obtain the initial features of each image in the easily separable image subset.

[0014] Each of the texture enhancement modules is used to perform texture enhancement on the corresponding difficult-to-distinguish image clusters to obtain the final texture features of each image in the difficult-to-distinguish image clusters;

[0015] The classifier is used to classify diseases based on the initial features of each image in the easily classifiable image subset and the final texture features of each image in each difficult-to-classify image cluster.

[0016] Acquire images of road surface defects to be tested, and use the trained road surface defect image classification model to obtain the defect classification results of the road surface defects images to be tested.

[0017] Furthermore, the enhancement of the defect features of the defective pavement images in the initial pavement image dataset to obtain an enhanced set of defective pavement images specifically involves:

[0018] Obtain the current damaged road surface image from the initial road surface image dataset;

[0019] The current damaged road surface image is processed into a grayscale image to obtain the current damaged road surface image.

[0020] Based on the grayscale image of the current damaged road surface, the current damaged road surface image is divided into image blocks to obtain the image block division result of the current damaged road surface image;

[0021] Based on the image block segmentation results of the current damaged road surface image, feature enhancement is performed to obtain the enhanced damaged road surface image of the current damaged road surface image;

[0022] The next damaged pavement image is enhanced with damage features until the enhanced damaged pavement images of all damaged pavement images in the initial pavement image dataset are obtained, resulting in an enhanced set of damaged pavement images.

[0023] Furthermore, the step of dividing the current damaged road surface image into image blocks based on the grayscale image of the current damaged road surface image to obtain the image block division result of the current damaged road surface image is as follows:

[0024] The target detection algorithm is used to identify targets in the current image of the damaged road surface to obtain the bounding box of the damaged area.

[0025] Set the outline magnification parameters; based on the outline magnification parameters, expand the disease area box outward to obtain the expanded disease area box;

[0026] The expanded diseased area bounding box is cut out from the grayscale image of the current diseased pavement image to obtain the background area without disease;

[0027] Based on the expanded defect area bounding box and the undamaged background area, the image block segmentation result of the current defective pavement image is obtained.

[0028] Furthermore, the step of performing feature enhancement based on the image block segmentation results of the current damaged road surface image to obtain an enhanced damaged road surface image specifically involves:

[0029] The grayscale values ​​of each pixel in the disease-free background area are reduced to obtain the processed disease-free background area; the grayscale values ​​of each pixel in the processed disease-free background area are:

[0030]

[0031] Where F(i) is the gray value of pixel i in the processed, disease-free background region; H i Z(H) represents the grayscale value of pixel i in the background area free of disease; i () represents a background area with a grayscale value of H where there are no defects. i The number of pixels; R is the total number of pixels in the background area without defects;

[0032] The processed, disease-free background area and the enlarged disease area frame are superimposed onto a single image to obtain the integrated image of the diseased road surface.

[0033] The integrated image of the damaged road surface is converted into an RGB image to obtain an enhanced image of the damaged road surface.

[0034] Furthermore, the expression for the cleaned damaged pavement images in the set of cleaned damaged pavement images is:

[0035]

[0036] Where X” is the cleaned image of the damaged pavement; f() is the binarization operation; g() is the inverse binarization operation; X' is the enhanced image of the damaged pavement after closing operation; S2 and S1 are both structuring elements; X is the enhanced image of the damaged pavement; ⊕ is the dilation operation; This is a corrosion operation.

[0037] Further, the step of dividing the cleaned pavement damage image set into easily separable image subsets and difficult-to-separate image subsets specifically involves: obtaining the initial features of each cleaned pavement damage image in the cleaned pavement damage image set, and treating the initial features of each cleaned pavement damage image as a cluster; obtaining discrete clusters, and assigning the cleaned pavement damage images belonging to the discrete clusters to the easily separable image subset; removing the clusters from each image in the easily separable image subset to obtain the remaining clusters; calculating the distance between each cluster in the remaining clusters; merging clusters that meet the merging conditions until the number of clusters is less than or equal to a preset number of clusters, resulting in several difficult-to-separate image clusters; the expression for the distance between the clusters is:

[0038]

[0039] Where D is the distance between clusters; min is the minimum value function; max is the maximum value function; d is the distance calculation function; V A V represents the initial feature vector of the image in cluster A; B This represents the initial feature vector of the image in cluster B; This is the average distance between the initial feature vectors of images in cluster A and cluster B.

[0040] Furthermore, the texture enhancement module includes a texture feature extractor, a semantic feature extractor, and a weighted fusion unit;

[0041] The texture feature extractor is used to extract local texture features of each image in a difficult-to-distinguish image cluster.

[0042] The semantic feature extractor is used to extract the semantic features of each image in a difficult-to-distinguish image cluster.

[0043] The weighted fusion unit is used to perform weighted fusion of the texture features and semantic features of each image in the difficult-to-distinguish image cluster to obtain the final texture features of each image in the difficult-to-distinguish image cluster.

[0044] Furthermore, the expression for the final texture feature is:

[0045] F T =Resnet18([F LBP ,F Con ,F Cor ,F ASM ,F Dis ,F Hom ,F Ent ])

[0046] F S =S(F GLCM ),S∈{Con,Cor,ASM,Dis,Hom,Ent}

[0047] Among them, F T ' represents the final texture features; ResNet18 is a pre-trained ResNet18 network; F LBP For local texture features; F Con For contrast characteristics; F Cor For correlation features; F ASM It is a second-order moment characteristic of the angle; F Dis For differential characteristics; F Hom It is a homogeneous characteristic; F Ent Entropy features; F S S represents the overall feature map; S represents the feature selection operation; F represents the feature map. GLCM For semantic features; Con is the contrast identifier; Cor is the correlation identifier; ASM is the second moment identifier; Dis is the difference identifier; Hom is the homogeneity identifier; Ent is the entropy identifier.

[0048] Furthermore, the classifier includes a general classification head for classifying diseases based on the initial features of each image in the easily classifiable image subset, and N special classification heads for classifying diseases based on the final texture features of each image in each difficult-to-classify image cluster.

[0049] Furthermore, the step of acquiring the pavement distress image to be tested and using the trained pavement distress image classification model to obtain the distress classification result of the pavement distress image to be tested specifically involves:

[0050] The process involves acquiring images of road surface defects to be tested, and using a defect feature enhancement module and a feature map cleaning module to obtain a cleaned image. The distances between the cleaned image and the easily separable subsets and difficult-to-separate image clusters defined in the feature map clustering module are calculated, and the smallest distance is selected as the input module. If the distance to the easily separable subset is the smallest, the cleaned image is input into the feature extraction module, and a standard classification head is used to obtain the defect classification result. Otherwise, the cleaned image is input into the texture enhancement module with the smallest distance, and a corresponding special classification head is used to obtain the defect classification result.

[0051] The beneficial effects of this invention are as follows: This method enhances the damage features by reducing the background information of the images in the training set, enabling the trained road damage image classification model to read more obvious damage features and thus achieve classification. The same operation is performed on the detection targets: first, the target box is obtained, and then the contrast with the background is enhanced to enhance the damage features. This process avoids the dependence of the image classification model on image quality. At the same time, during the training process, a coarse classification is first performed on the training set, and different classification channels are divided in the road damage image classification model so that each channel achieves the optimal classification accuracy in the corresponding image cluster, thus avoiding the errors caused by training with a large amount of data. Attached Figure Description

[0052] Figure 1 This is a flowchart of the method of the present invention.

[0053] Figure 2 This is a model framework diagram of the present invention. Detailed Implementation

[0054] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0055] like Figure 1 and Figure 2 As shown, in one embodiment of the present invention, a pavement distress analysis method based on distress feature enhancement and image classification model includes:

[0056] Obtain an initial dataset of road surface images; the initial dataset of road surface images includes images of damaged road surfaces and information on road surface defects.

[0057] A road surface defect image classification model is constructed; the road surface defect image classification model is trained using an initial dataset of road surface images to obtain a trained road surface defect image classification model; the road surface defect image classification model includes a defect feature enhancement module, a feature map cleaning module, a feature map clustering module, a feature extraction module, several texture enhancement modules, and a classifier;

[0058] The defect feature enhancement module is used to enhance the defect features of defective pavement images in the initial pavement image dataset to obtain an enhanced set of defective pavement images.

[0059] The feature map cleaning module is used to remove interfering features from the enhanced set of damaged road surface images to obtain a cleaned set of damaged road surface images.

[0060] The feature map clustering module is used to divide the cleaned image set of damaged road surfaces into easily separable image subsets and difficult-to-separate image subsets; the difficult-to-separate image subsets include several difficult-to-separate image clusters.

[0061] The feature extraction module is used to extract features from each image in the easily separable image subset to obtain the initial features of each image in the easily separable image subset.

[0062] Each of the texture enhancement modules is used to perform texture enhancement on the corresponding difficult-to-distinguish image clusters to obtain the final texture features of each image in the difficult-to-distinguish image clusters;

[0063] The classifier is used to classify diseases based on the initial features of each image in the easily classifiable image subset and the final texture features of each image in each difficult-to-classify image cluster.

[0064] Acquire images of road surface defects to be tested, and use the trained road surface defect image classification model to obtain the defect classification results of the road surface defects images to be tested.

[0065] In this embodiment, the feature extraction module identifies road surface defects based on the Transformer model combined with traditional convolution. Traditional convolutional layers are used to extract features from the preprocessed road surface image. Convolutional layers can capture local features of the image, including texture and shape. By adjusting the size and number of convolutional kernels, features at different scales can be extracted. A self-attention mechanism is used to analyze the output of the feature fusion layer using a Transformer layer. Through this self-attention mechanism, the model can automatically learn important regions in the image and assign them greater attention. This layer can capture global features of the image. The Transformer can divide the road surface image into multiple small blocks for feature extraction and transformation, and use Patch Embedding to embed each small block into a vector; simultaneously, it retains the depth and width of the Transformer, enabling it to better capture the image's features and patterns.

[0066] In Convolutional Neural Networks (CNNs), convolution operations excel at extracting local features, but they have limitations in capturing global feature representations. In Vision Transformers, cascaded self-attention modules can capture long-range feature dependencies, but long-range attention mechanisms ignore the details of local features. This project, based on a multi-layered Transformer, inserts traditional convolutions between network layers for fine-grained feature extraction. CNNs at lower layers are used to extract local image features, while Transformers at higher layers are used to capture global features. By fusing features at different scales, the aim is to hierarchically extract local and global features from images of damaged road surfaces using a convolution + global attention approach, thereby improving model performance and reducing computational cost.

[0067] The enhancement of the defect features of the defective pavement images in the initial dataset of pavement images to obtain an enhanced set of defective pavement images is specifically as follows:

[0068] Obtain the current damaged road surface image from the initial road surface image dataset;

[0069] The current damaged road surface image is processed into a grayscale image to obtain the current damaged road surface image.

[0070] Based on the grayscale image of the current damaged road surface, the current damaged road surface image is divided into image blocks to obtain the image block division result of the current damaged road surface image;

[0071] Based on the image block segmentation results of the current damaged road surface image, feature enhancement is performed to obtain the enhanced damaged road surface image of the current damaged road surface image;

[0072] The next damaged pavement image is enhanced with damage features until the enhanced damaged pavement images of all damaged pavement images in the initial pavement image dataset are obtained, resulting in an enhanced set of damaged pavement images.

[0073] The process involves dividing the current damaged road surface image into image blocks based on its grayscale value, resulting in the image block division of the current damaged road surface image. Specifically:

[0074] The target detection algorithm is used to identify targets in the current image of the damaged road surface to obtain the bounding box of the damaged area.

[0075] Set the outline magnification parameters; based on the outline magnification parameters, expand the disease area box outward to obtain the expanded disease area box;

[0076] The expanded diseased area bounding box is cut out from the grayscale image of the current diseased pavement image to obtain the background area without disease;

[0077] Based on the expanded defect area bounding box and the undamaged background area, the image block segmentation result of the current defective pavement image is obtained.

[0078] The step of performing feature enhancement based on the image block segmentation results of the current damaged road surface image to obtain an enhanced damaged road surface image is as follows:

[0079] The grayscale values ​​of each pixel in the disease-free background area are reduced to obtain the processed disease-free background area; the grayscale values ​​of each pixel in the processed disease-free background area are:

[0080]

[0081] Where F(i) is the gray value of pixel i in the processed, disease-free background region; H i Z(H) represents the grayscale value of pixel i in the background area free of disease; i () represents a background area with a grayscale value of H where there are no defects. i The number of pixels; R is the total number of pixels in the background area without defects;

[0082] The processed, disease-free background area and the enlarged disease area frame are superimposed onto a single image to obtain the integrated image of the diseased road surface.

[0083] The integrated image of the damaged road surface is converted into an RGB image to obtain an enhanced image of the damaged road surface.

[0084] In this embodiment, the contour magnification parameter is set to expand the original disease area box outward as a whole. This can be used to avoid the disease edge being cut off during subsequent grayscale reduction, which would lead to errors in the extraction of road surface disease features.

[0085] In this embodiment, the grayscale of the background area is reduced proportionally, which can highlight the characteristics of road surface defects while retaining certain background features.

[0086] In this embodiment, grayscale reduction is performed according to the numerical ratio of grayscale values, which can preserve the global information of the background to the greatest extent, while highlighting the characteristics of road surface defects.

[0087] The expression for the cleaned damaged road surface images in the set of cleaned damaged road surface images is:

[0088]

[0089] Where X” is the cleaned image of the damaged pavement; f() is the binarization operation; g() is the inverse binarization operation; X' is the enhanced image of the damaged pavement after closing operation; S2 and S1 are both structuring elements; X is the enhanced image of the damaged pavement; ⊕ is the dilation operation; This is a corrosion operation.

[0090] The process of dividing the cleaned pavement damage image set into easily separable and difficult-to-separate image subsets involves: obtaining the initial features of each cleaned pavement damage image in the cleaned pavement damage image set, and treating each initial feature as a cluster; obtaining discrete clusters, and assigning the cleaned pavement damage images belonging to the discrete clusters to the easily separable image subset; removing the clusters from each image in the easily separable image subset to obtain the remaining clusters; calculating the distance between each cluster in the remaining clusters; merging clusters that meet the merging conditions until the number of clusters is less than or equal to a preset number of clusters, resulting in several difficult-to-separate image clusters; the expression for the distance between the clusters is:

[0091]

[0092] Where D is the distance between clusters; min is the minimum value function; max is the maximum value function; d is the distance calculation function; V A V represents the initial feature vector of the image in cluster A; B This represents the initial feature vector of the image in cluster B; This is the average distance between the initial feature vectors of images in cluster A and cluster B.

[0093] The texture enhancement module includes a texture feature extractor, a semantic feature extractor, and a weighted fusion unit;

[0094] The texture feature extractor is used to extract local texture features of each image in a difficult-to-distinguish image cluster.

[0095] The semantic feature extractor is used to extract the semantic features of each image in a difficult-to-distinguish image cluster.

[0096] The weighted fusion unit is used to perform weighted fusion of the texture features and semantic features of each image in the difficult-to-distinguish image cluster to obtain the final texture features of each image in the difficult-to-distinguish image cluster.

[0097] The expression for the final texture feature is:

[0098] F T =Resnet18([F LBP ,F Con ,F Cor ,F ASM ,F Dis ,F Hom ,F Ent ])

[0099] F S =S(F GLCM ),S∈{Con,Cor,ASM,Dis,Hom,Ent}

[0100] Among them, F T ' represents the final texture features; ResNet18 is a pre-trained ResNet18 network; F LBP For local texture features; F Con For contrast characteristics; F Cor For correlation features; F ASM It is a second-order moment characteristic of the angle; F Dis For differential characteristics; F Hom It is a homogeneous characteristic; F Ent Entropy features; F S S represents the overall feature map; S represents the feature selection operation; F represents the feature map. GLCM For semantic features; Con is the contrast identifier; Cor is the correlation identifier; ASM is the second moment identifier; Dis is the difference identifier; Hom is the homogeneity identifier; Ent is the entropy identifier.

[0101] The classifier includes a general classification head for classifying diseases based on the initial features of each image in an easily classifiable subset, and N special classification heads for classifying diseases based on the final texture features of each image in each difficult-to-classify image cluster.

[0102] The process of acquiring the pavement distress image to be tested and using the trained pavement distress image classification model to obtain the distress classification result of the pavement distress image to be tested is as follows:

[0103] The process involves acquiring images of road surface defects to be tested, and using a defect feature enhancement module and a feature map cleaning module to obtain a cleaned image. The distances between the cleaned image and the easily separable subsets and difficult-to-separate image clusters defined in the feature map clustering module are calculated, and the smallest distance is selected as the input module. If the distance to the easily separable subset is the smallest, the cleaned image is input into the feature extraction module, and a standard classification head is used to obtain the defect classification result. Otherwise, the cleaned image is input into the texture enhancement module with the smallest distance, and a corresponding special classification head is used to obtain the defect classification result.

Claims

1. A method for analyzing pavement defects based on defect feature enhancement and image classification models, characterized in that, include: Obtain an initial dataset of road surface images; the initial dataset of road surface images includes images of damaged road surfaces and information on road surface defects. Construct a road surface distress image classification model; A road surface defect image classification model is trained using an initial dataset of road surface images to obtain the trained road surface defect image classification model. The road surface defect image classification model includes a defect feature enhancement module, a feature map cleaning module, a feature map clustering module, a feature extraction module, several texture enhancement modules, and a classifier. The defect feature enhancement module is used to enhance the defect features of defective pavement images in the initial pavement image dataset to obtain an enhanced set of defective pavement images. The feature map cleaning module is used to remove interfering features from the enhanced set of damaged road surface images to obtain a cleaned set of damaged road surface images. The feature map clustering module is used to divide the cleaned image set of damaged road surfaces into easily separable image subsets and difficult-to-separate image subsets; the difficult-to-separate image subsets include several difficult-to-separate image clusters. The feature extraction module is used to extract features from each image in the easily separable image subset to obtain the initial features of each image in the easily separable image subset. Each of the texture enhancement modules is used to perform texture enhancement on the corresponding difficult-to-distinguish image clusters to obtain the final texture features of each image in the difficult-to-distinguish image clusters; The classifier is used to classify diseases based on the initial features of each image in the easily classifiable image subset and the final texture features of each image in each difficult-to-classify image cluster. Acquire images of road surface defects to be tested, and use the trained road surface defect image classification model to obtain the defect classification results of the road surface defects images to be tested.

2. The pavement distress analysis method based on distress feature enhancement and image classification model according to claim 1, characterized in that, The enhancement of the defect features of the defective pavement images in the initial dataset of pavement images to obtain an enhanced set of defective pavement images is specifically as follows: Obtain the current damaged road surface image from the initial road surface image dataset; The current damaged road surface image is processed into a grayscale image to obtain the current damaged road surface image. Based on the grayscale image of the current damaged road surface, the current damaged road surface image is divided into image blocks to obtain the image block division result of the current damaged road surface image; Based on the image block segmentation results of the current damaged road surface image, feature enhancement is performed to obtain the enhanced damaged road surface image of the current damaged road surface image; The next damaged pavement image is enhanced with damage features until the enhanced damaged pavement images of all damaged pavement images in the initial pavement image dataset are obtained, resulting in an enhanced set of damaged pavement images.

3. The pavement distress analysis method based on distress feature enhancement and image classification model according to claim 2, characterized in that, The process involves dividing the current damaged road surface image into image blocks based on its grayscale value, resulting in the image block division of the current damaged road surface image. Specifically: The target detection algorithm is used to identify targets in the current image of the damaged road surface to obtain the bounding box of the damaged area. Set the outline magnification parameters; based on the outline magnification parameters, expand the disease area box outward to obtain the expanded disease area box; The expanded diseased area bounding box is cut out from the grayscale image of the current diseased pavement image to obtain the background area without disease; Based on the expanded defect area bounding box and the undamaged background area, the image block segmentation result of the current defective pavement image is obtained.

4. The pavement distress analysis method based on distress feature enhancement and image classification model according to claim 3, characterized in that, The step of performing feature enhancement based on the image block segmentation results of the current damaged road surface image to obtain an enhanced damaged road surface image is as follows: The grayscale values ​​of each pixel in the disease-free background area are reduced to obtain the processed disease-free background area; the grayscale values ​​of each pixel in the processed disease-free background area are: Where F(i) is the gray value of pixel i in the processed, disease-free background region; H i Z(H) represents the grayscale value of pixel i in the background area free of disease; i () represents a background area with a grayscale value of H where there are no defects. i The number of pixels; R is the total number of pixels in the background area without defects; The processed, disease-free background area and the enlarged disease area frame are superimposed onto a single image to obtain the integrated image of the diseased road surface. The integrated image of the damaged road surface is converted into an RGB image to obtain an enhanced image of the damaged road surface.

5. The pavement distress analysis method based on distress feature enhancement and image classification model according to claim 1, characterized in that, The expression for the cleaned damaged road surface images in the set of cleaned damaged road surface images is: Where X” is the cleaned image of the damaged pavement; f() is the binarization operation; g() is the inverse binarization operation; X' is the enhanced image of the damaged pavement after closing operation; S2 and S1 are both structuring elements; X is the enhanced image of the damaged pavement. For expansion operation; This is a corrosion operation.

6. The pavement distress analysis method based on distress feature enhancement and image classification model according to claim 1, characterized in that, The step of dividing the cleaned damaged road surface image set into easily separable image subsets and difficult-to-separate image subsets specifically involves: obtaining the initial features of each cleaned damaged road surface image in the cleaned damaged road surface image set, and treating the initial features of each cleaned damaged road surface image as a cluster class; obtaining discrete cluster classes, and classifying the cleaned damaged road surface images to which the discrete cluster classes belong into the easily separable image subset; removing the cluster classes of each image in the easily separable image subset to obtain the remaining cluster classes; The distance between each cluster in the remaining clusters is calculated; clusters that meet the merging conditions are merged until the number of clusters is less than or equal to the preset number of clusters, resulting in several hard-to-distinguish image clusters; the expression for the distance between the clusters is: Where D is the distance between clusters; min is the minimum value function; max is the maximum value function; d is the distance calculation function; V A V represents the initial feature vector of the image in cluster A; B This represents the initial feature vector of the image in cluster B; This is the average distance between the initial feature vectors of images in cluster A and cluster B.

7. The pavement distress analysis method based on distress feature enhancement and image classification model according to claim 1, characterized in that, The texture enhancement module includes a texture feature extractor, a semantic feature extractor, and a weighted fusion unit; The texture feature extractor is used to extract local texture features of each image in a difficult-to-distinguish image cluster. The semantic feature extractor is used to extract the semantic features of each image in a difficult-to-distinguish image cluster. The weighted fusion unit is used to perform weighted fusion of the texture features and semantic features of each image in the difficult-to-distinguish image cluster to obtain the final texture features of each image in the difficult-to-distinguish image cluster.

8. The pavement distress analysis method based on distress feature enhancement and image classification model according to claim 7, characterized in that, The expression for the final texture feature is: F T '=Resnet18([F LBP ,F Con ,F Cor ,F ASM ,F Dis ,F Hom ,F Ent ]) F S =S(F GLCM ),S∈{Con,Cor,ASM,Dis,Hom,Ent} Among them, F T ' represents the final texture features; ResNet18 is a pre-trained ResNet18 network; F LBP For local texture features; F Con For contrast characteristics; F Cor For correlation features; F ASM It is a second-order moment characteristic of the angle; F Dis For differential characteristics; F Hom It is a homogeneous characteristic; F Ent Entropy features; F S S represents the overall feature map; S represents the feature selection operation; F represents the feature map. GLCM For semantic features; Con is the contrast identifier; Cor is the correlation identifier; ASM is the second moment identifier; Dis is the difference identifier; Hom is the homogeneity identifier; Ent is the entropy identifier.

9. The pavement distress analysis method based on distress feature enhancement and image classification model according to claim 1, characterized in that, The classifier includes a general classification head for classifying diseases based on the initial features of each image in an easily classifiable subset, and N special classification heads for classifying diseases based on the final texture features of each image in each difficult-to-classify image cluster.

10. The pavement distress analysis method based on distress feature enhancement and image classification model according to claim 1, characterized in that, The process of acquiring the pavement distress image to be tested and using the trained pavement distress image classification model to obtain the distress classification result of the pavement distress image to be tested is as follows: The process involves acquiring images of road surface defects to be tested, using a defect feature enhancement module and a feature map cleaning module to obtain cleaned images of the road surface defects to be tested, calculating the distances between the cleaned images of the road surface defects to be tested and the easily separable image subsets and the difficult-to-separate image clusters in the feature map clustering module, and selecting the one with the smallest distance as the subsequent input module. If the distance to the easily separable subset of images is the smallest, the cleaned pavement defect image to be tested is input into the feature extraction module, and the defect classification result of the pavement defect image to be tested is obtained using a common classification head; otherwise, the cleaned pavement defect image to be tested is input into the texture enhancement module with the smallest distance, and the defect classification result of the pavement defect image to be tested is obtained using the corresponding special classification head.

Images