Road surface crack real-time detection and classification method adopting digital twinborn technology

Abstract

The invention discloses a pavement crack real-time detection and classification method adopting a digital twin technology, and belongs to the technical field of urban intelligent operation and maintenance. In order to detect and classify pavement cracks in real time, the method comprises the following steps: performing image quality enhancement processing on an acquired image matrix to obtain a quality-enhanced image matrix; carrying out image pixel updating on the quality-enhanced image matrix to obtain a pixel-updated image matrix, and storing the pixel-updated image matrix as a pixel-updated pavement image; all images including cracks, spalling, pit slots and upheaval diseases are selected and marked, and a data set for real-time pavement crack detection is obtained; a pavement crack rapid identification method integrating multi-model features is constructed; constructing a road digital twinborn model based on a pavement crack rapid identification method integrating multi-model features; and establishing a pavement crack classification method based on the obtained road digital twin model. The pertinence and effectiveness of road repair are improved, the service life of the road is prolonged, and the total cost of road maintenance is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of urban intelligent operation and maintenance, and specifically relates to a real-time detection and classification method for pavement cracks using digital twin technology. Background Art

[0002] Pavement cracks have a significant impact on road use and traffic safety. First, cracks can reduce the overall structural integrity and load-bearing capacity of the road surface, making the road more vulnerable to further damage. Second, the presence of cracks affects the smoothness of the road surface. Drivers may feel bumps and vibrations during driving, reducing driving comfort. In addition, cracks can become infiltration channels for water and other corrosive substances, which can further damage the roadbed and pavement materials, especially during freeze-thaw cycles, when the expansion of water after freezing will aggravate the expansion of cracks.

[0003] The classification of pavement cracks is a key step in road maintenance and management, as different types of cracks may indicate different structural problems and failure mechanisms. An effective crack classification method can help road engineers accurately diagnose the causes of cracks and take more targeted repair measures. Systematic crack classification helps optimize road maintenance plans and budget allocations, ensuring that resources are effectively used for maintenance activities that are most needed. Through scientific classification, the efficiency and effectiveness of road maintenance can be improved, the service life of roads can be extended, and maintenance costs can be reduced.

[0004] The invention patent with application number 202210813650.2 and invention name "Method for detecting and identifying crack geometric properties in highway pavement images based on deep convolutional neural networks" solves the limitations of traditional crack detection methods in crack expression by introducing a crack expression method based on geometric properties, and can accurately describe the position, direction and length of cracks in the image. By constructing a deep convolutional neural network and designing a new loss function, this method can efficiently detect and identify the geometric properties of cracks, and classify cracks into types such as horizontal stripes, longitudinal stripes or cracks. However, in the above-mentioned studies, the classification method of pavement cracks was not considered, which would lead to the inability to accurately diagnose the causes of cracks, and thus take inappropriate maintenance measures, resulting in waste of resources and poor maintenance effects, ultimately affecting the service life and safety of the road. Summary of the invention

[0005] The problem to be solved by the present invention is real-time detection and classification of pavement cracks, and a real-time detection and classification method for pavement cracks using digital twin technology is proposed.

[0006] To achieve the above object, the present invention is implemented through the following technical solutions: A real-time detection and classification method for pavement cracks using digital twin technology includes the following steps: S1. Use a high-definition camera on a vehicle to collect road images and save them as an image matrix; S2. Perform image quality enhancement processing on the image matrix acquired in step S1 to obtain an image matrix with enhanced quality; S3. Performing image pixel updates on the quality enhanced image matrix obtained in step S2 to obtain a pixel updated image matrix, which is saved as a pixel updated road surface image; S4. For the pixel updated road image obtained in step S3, all images including cracks, peeling, potholes, and bagging diseases are selected and annotated to obtain a data set for real-time detection of road cracks; S5. Construct a fast pavement crack identification method integrating multi-model features; S6. Constructing a road digital twin model based on a road crack rapid identification method integrating multi-model features in step S5; S7. Based on the road digital twin model obtained in step S6, a pavement crack classification method is established.

[0007] Furthermore, the road surface image collected in step S1 is denoted as ph, and the corresponding image matrix is ​​denoted as PH .

[0008] Furthermore, the specific implementation method of step S2 includes the following steps: S2.1. Constructing the 5th-order denoising kernel function matrix N , the calculation formula is: ; in, is the element in the kernel function matrix, is the row number of the 5th-order denoising kernel function matrix, is the column number of the 5th-order denoising kernel function matrix; ; in, is the standard deviation of all elements in the kernel function; S2.2. Normalize all elements in the kernel function matrix to obtain the normalized denoising kernel function matrix , the calculation formula is: ; S2.3. Perform denoising kernel function processing on the image matrix collected in step S1 to obtain the image matrix processed by denoising kernel function P , the calculation formula is: .

[0009] Furthermore, the specific implementation method of step S3 includes the following steps: S3.1. Divide the pixel grayscale value 0~255 into 16 grayscale intervals, denoted as , is the gray level number, =1,2,3,…,16; S3.2. The image matrix processed by the denoising kernel function obtained in step S2 is divided into N equal parts. The image matrix after equal division is recorded as P num, num is the number of the image matrix after equal division, num=1,2,…,N; S3.3. Count each gray level interval in the equally divided image matrix The frequency of the corresponding pixel ; S3.4. Calculate each gray level in Pnum The corresponding grayscale frequency cumulative value , the calculation formula is: ; in, is less than or equal to The gray level number, =0,1,2,…, ; S3.5. Construct a pixel value update method, the calculation formula is: ; in, is the minimum non-zero value among all frequency accumulation values, is the image matrix after equal division P The total number of pixels corresponding to num, is the maximum value of the gray level, Grayscale The corresponding pixel update value; S3.6. Perform pixel value update processing on all equally divided image matrices to obtain pixel-updated image matrices, and save them as pixel-updated road surface images.

[0010] Furthermore, the specific implementation method of step S4 includes the following steps: S4.1. For the pixel-updated road surface image obtained in step S3, select all images containing cracks, peeling, potholes, and bulges; S4.2. By manually labeling the crack area and the defects of spalling, potholes, and lumps in the image, the cracks are marked as Crack, and the other three defects are marked as other1, other2, and other3, respectively, to construct a dataset for real-time detection of pavement cracks. S4.3. The data set for real-time detection of pavement cracks obtained in step S4.2 is divided into a training set and a test set in a ratio of 8:2.

[0011] Furthermore, the specific implementation method of step S5 includes the following steps: S5.1. Construct a first deep learning model A; S5.1.1. The first deep learning model A consists of 1 input layer, 4 convolutional layers, 5 activation layers, 3 pooling layers, 1 flatten layer, 1 fully connected layer, and 1 output layer; S5.1.2. Input the training set obtained in step S4 into the first deep learning model A, calculate the weights and biases of the neural network units through the error back propagation algorithm, complete the training of the first deep learning model A, and obtain the trained first deep learning model A. The calculation formula of the loss function of the first deep learning model A is: ; in, is the loss function of the first deep learning model A, is the total number of data samples in the first deep learning model A, For the The true labels of samples; The model predicts The probability that a sample is class 1; S5.1.3. The output values ​​of the four output unit nodes corresponding to the first deep learning model A are , , , ; Perform initial conversion on the node output values ​​of the 4 output units to obtain the output value The corresponding probability value The calculation formula is: ; Then the node output values ​​of the four output units are further optimized and converted to obtain the output value The corresponding enhancement probability value The calculation formula is: ; in, G is the enhancement coefficient, and then the actual disease corresponding to each node is marked as ; S5.1.4. Adoption And the training set, combined with the error chain derivative rule, train the first deep learning model A. After the training is completed, the output values ​​of the corresponding four types of diseases are obtained, which are recorded as , , , ; S5.2. Constructing a second deep learning model B; S5.2.1. The second deep learning model B consists of 1 input layer, 8 convolutional layers, 9 activation layers, 5 pooling layers, 1 Flatten layer, 1 fully connected layer, and 1 output layer; S5.2.2. Input the training set obtained in step S4 into the second deep learning model B, calculate the weights and biases of the neural network units through the error back propagation algorithm, complete the training of the second deep learning model B, and obtain the trained second deep learning model B. The calculation formula of the loss function of the second deep learning model B is: ; in, is the loss function of the second deep learning model B; is the total number of data samples in the second deep learning model B; For the The true labels of samples, The second deep learning model B predicts The probability that a sample is class 1; S5.2.3. The node output values ​​of the four output units corresponding to the second deep learning model B are b 1. b 2. b 3. b 4. Perform initial conversion on the node output values ​​of the four output units to obtain the output values The corresponding probability value The calculation formula is: ; Then the node output values ​​of the four output units are further optimized and converted to obtain the output value b i The corresponding enhancement probability value The calculation formula is: ; S5.2.4. Adoption The training set obtained in step S4 is combined with the error chain derivation rule to train the second deep learning model B. After the training is completed, the disease output value corresponding to the second deep learning model B is obtained as follows: , among the 4 nodes in the output layer, the actual disease corresponding to each node is marked as , =1,2,3,4; S5.3. Construct a third deep learning model C; S5.3.1. The third deep learning model C consists of 1 input layer, 12 convolutional layers, 13 activation layers, 6 pooling layers, 1 Flatten layer, 1 fully connected layer, and 1 output layer; S5.3.2. Input the training set obtained in step S4 into the third deep learning model C, calculate the weights and biases of the neural network units through the error back propagation algorithm, complete the training of the third deep learning model C, and obtain the trained third deep learning model C; The calculation formula of the loss function of the third deep learning model C is: ; in, is the loss function of the third deep learning model C; is the total number of data samples in the third deep learning model C; For the The true labels of samples, The third deep learning model C predicts The probability that a sample is class 1; S5.3.3. The node output values ​​of the four output units corresponding to the third deep learning model C are , the node output values ​​of the four output units are initially converted to obtain The corresponding probability value The calculation formula is: ; Then the node output values ​​of the four output units are further optimized and converted to obtain the output value The corresponding enhancement probability value The calculation formula is: ; in, G is the enhancement coefficient, and then the actual disease corresponding to each node is marked as ; S5.3.4. Adoption The training set obtained in step S4 is combined with the error chain derivation rule to train the third deep learning model C. After the training is completed, the output values ​​of the four types of diseases corresponding to the third deep learning model C are obtained as follows: .

[0012] S5.4. Based on and , establish the loss function between the second deep learning model B and the output layer of the first deep learning model A, and the calculation formula is: ; based on and , construct the output value The corresponding probability value The loss function of the first deep learning model A , the calculation formula is: ; based on and , establish the relationship loss function between the second deep learning model B and the output layer of the first deep learning model A , the calculation formula is: , in, for The weight coefficient of for The weight coefficient of S5.5. Establish the relationship loss function between the output layer of the third deep learning model C and the second deep learning model B , the expression is: ; in, for The weight coefficient of for The weight coefficient of Based on and The loss function of Based on and The loss function of The calculation formula is: ; The calculation formula is: ; S5.6. Establish the relationship loss function between the output layer of the third deep learning model C and the first deep learning model A , the expression is: ; in, for The weight coefficient of for The weight coefficient of Based on and The loss function is Based on and The loss function of The calculation formula is: ; The calculation formula is: ; S5.7. Define the overall loss function corresponding to the first deep learning model A, the second deep learning model B, and the third deep learning model C , the calculation formula is: ; S5.8. Using the data set for rutting recognition obtained in step S4, based on the chain rule and through error back propagation, the first deep learning model A, the second deep learning model B, and the third deep learning model C are trained. When the overall loss function obtained in step S5.7 is When is the minimum, the corresponding model parameters are optimal, and the first deep learning A after training is the pavement crack rapid identification model with comprehensive multi-model features that is finally used.

[0013] Furthermore, the specific implementation method of step S6 includes the following steps: S6.1. Export the pavement crack rapid identification model with integrated multi-model features obtained in step S5 as a deployment file for independent services, including one of a pt file, a pth file, a json file, and a txt file; S6.2. Construct a road information model; S6.2.1. Collect information on the location, length, width, and layout of roads; S6.2.2. The data obtained in step S6.2.1 is stored in a database, and a basic three-dimensional model of the road is constructed, wherein the basic three-dimensional model of the road includes geometric information of the road; S6.2.3. Based on step S6.2.2, material properties, design specifications, and historical maintenance data are added to the basic three-dimensional model of the road to obtain a road information model; S6.2.4. Export the road information model to a format file supported by the web application, including a json file or a js file; S6.3. In the digital twin platform, import the deployment file of the pavement crack rapid identification model with integrated multi-model features obtained in step S6.1, and import the road information model format file obtained in step S7.2; S6.4. Use Python to construct an API interface between image data and a road surface crack rapid identification model with integrated multi-model features for intelligent rutting identification. Use Python to construct an API interface between a road surface crack rapid identification model with integrated multi-model features and a road information model to complete the construction of a road digital twin model.

[0014] Furthermore, the specific implementation method of step S7 includes the following steps: S7.1. Use a high-definition camera on a vehicle to collect road images and upload them to the road digital twin model constructed in step S6, intelligently identify crack areas in the road images, and select all images containing road cracks; S7.2. For all pavement crack regions of the image of pavement cracks identified in step S7.1, obtain the length of the cracks by on-site verification and measurement. CD , Width data KD ; Then, collect crack information at different times, compare crack length and width data, and calculate length development speed CF , Width Development Speed KF ; S7.3. Construct a crack criteria matrix based on the length of the cracks CD ,width KD , length development speed CF , Width Development Speed KF ,quantity SL , a road pavement crack state classification standard is established, and the crack state is divided into NL categories. The length standard value, width standard value, length growth speed standard value, width growth speed standard value, and quantity standard value are recorded as , ; Establish a crack standard matrix based on the standard value of each type of crack, Crack-like standard matrix The expression is: ; S7.3. Classify pavement cracks; S7.3.1. Collect road surface images for the NW road, use the road digital twin model to identify images containing road surface cracks, and obtain the first l The size information of the road cracks on the road, including length, width, and number, is collected and calculated at different times to calculate the length and width growth rates of the cracks. On this basis, the first l Crack matrix of the road , the expression is: ; S7.3.2. Calculate the l The crack matrix of the first road is Crack Criteria Matrix The distance between , the calculation formula is: ; in, is the weight coefficient between the crack length and the standard value of length, is the weight coefficient between the crack width and the standard value of the width, is the weight coefficient between the crack length development speed and the standard value of length development speed, is the weight coefficient between the crack width development speed and the standard value of the width development speed, is the weight coefficient between the number of cracks and the standard value of the number; S7.3.3. Calculate the l The distance between the crack matrix of the road and all the crack standard matrices is calculated, and the maximum value is found. The crack standard corresponding to the maximum value is l Classification of road pavement crack states corresponding to the roads; S7.3.4. Based on the methods of steps S7.3.2 and S7.3.3, the distances between all crack standard matrices corresponding to all NW roads are calculated to complete the classification of road pavement crack states of all roads.

[0015] Beneficial effects of the present invention: The present invention discloses a real-time pavement crack detection and classification method using digital twin technology, and proposes an image quality enhancement method to reduce the impact of noise on image quality.

[0016] The present invention discloses a real-time pavement crack detection and classification method using digital twin technology, and proposes a pavement crack image pixel update method to ensure that the image has a good contrast value, which is more helpful to extract crack features from the image and optimize the image contrast value.

[0017] The real-time detection and classification method for pavement cracks using digital twin technology described in the present invention reduces the complexity of the model and improves the recognition speed while ensuring the accuracy of pavement crack recognition.

[0018] The present invention discloses a real-time pavement crack detection and classification method using digital twin technology, which realizes the construction of a digital twin model and the accurate updating of crack areas.

[0019] The real-time detection and classification method for pavement cracks using digital twin technology described in the present invention can improve the pertinence and effectiveness of road repair, extend the service life of roads, and reduce the total cost of road maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flow chart of a real-time detection and classification method for pavement cracks using digital twin technology described in the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the present invention more clear, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the specific embodiments described are only part of the embodiments of the present invention, rather than all of the specific embodiments. The components of the specific embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations, and the present invention can also have other embodiments.

[0022] Therefore, the following detailed description of the specific embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents the selected specific embodiments of the present invention. Based on the specific embodiments of the present invention, all other specific embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0023] In order to further understand the content, features and effects of the present invention, the following specific implementation methods are given as examples, and the attached Figure 1 The detailed instructions are as follows:

[0024] Embodiment 1: A real-time detection and classification method for pavement cracks using digital twin technology includes the following steps: S1. Use a high-definition camera on a vehicle to collect road images and save them as an image matrix; Furthermore, the road surface image collected in step S1 is denoted as ph, and the corresponding image matrix is ​​denoted as PH .

[0025] The collected road surface images may contain noise due to a variety of external and internal factors. External factors include undesirable environmental conditions such as wind, rain, snow, and lighting changes, which can produce various visual disturbances on the image. Internal factors involve the acquisition equipment itself, such as electronic noise, thermal noise, quantization noise of the sensor, and stray light from the device's lens, which can introduce undesirable pixel variations during the image capture process. In addition, noise may also be introduced during image transmission and compression.

[0026] Noise can degrade image quality, mask or change key features in the image, and make it difficult for recognition algorithms to accurately detect and classify objects in the image. In the case of pavement crack detection, noise may blur crack boundaries or generate pseudo features similar to cracks, leading to false or missed detections. Noise also increases the computational complexity of image processing algorithms because the algorithms require additional steps to distinguish between real signals and noise, which consumes more computing resources and reduces processing speed, so image quality enhancement is required.

[0027] S2. Perform image quality enhancement processing on the image matrix acquired in step S1 to obtain an image matrix with enhanced quality; The specific implementation method of step S2 includes the following steps: S2.1. Constructing the 5th-order denoising kernel function matrix N , the calculation formula is: ; in, is the element in the kernel function matrix, is the row number of the 5th-order denoising kernel function matrix, is the column number of the 5th-order denoising kernel function matrix; ; in, is the standard deviation of all elements in the kernel function; S2.2. Normalize all elements in the kernel function matrix to obtain the normalized denoising kernel function matrix , the calculation formula is: ; S2.3. Perform denoising kernel function processing on the image matrix collected in step S1 to obtain the image matrix processed by denoising kernel function P , the calculation formula is: ; Affected by factors such as the acquisition environment, equipment, and driving status, it is often difficult for road images to maintain a good contrast during the acquisition process, which will directly affect the accuracy of target recognition. Unstable lighting conditions, equipment performance limitations, and camera instability caused by vehicle movement may all cause the collected images to have low contrast, making it difficult for key features such as road cracks to be clearly distinguished by the recognition algorithm, resulting in a decrease in recognition effect. In order to improve recognition efficiency, an image pixel update method is proposed.

[0028] S3. Performing image pixel updates on the quality enhanced image matrix obtained in step S2 to obtain a pixel updated image matrix, which is saved as a pixel updated road surface image; Furthermore, the specific implementation method of step S3 includes the following steps: S3.1. Divide the pixel grayscale value 0~255 into 16 grayscale intervals, denoted as , is the gray level number, =1,2,3,…,16; S3.2. The image matrix processed by the denoising kernel function obtained in step S2 is divided into N equal parts. The image matrix after equal division is recorded as P num, num is the number of the image matrix after equal division, num=1,2,…,N; S3.3. Count each gray level interval in the equally divided image matrix The frequency of the corresponding pixel ; S3.4. Calculate each gray level in Pnum The corresponding grayscale frequency cumulative value , the calculation formula is: ; in, is less than or equal to The gray level number, =0,1,2,…, ; S3.5. Construct a pixel value update method, the calculation formula is: ; in, is the minimum non-zero value among all frequency accumulation values, is the image matrix after equal division P The total number of pixels corresponding to num, is the maximum value of the gray level, Grayscale The corresponding pixel update value; S3.6. Perform pixel value update processing on all equally divided image matrices to obtain pixel-updated image matrices, and save them as pixel-updated road surface images.

[0029] S4. For the pixel updated road image obtained in step S3, all images including cracks, peeling, potholes, and bagging diseases are selected and annotated to obtain a data set for real-time detection of road cracks; Furthermore, the specific implementation method of step S4 includes the following steps: S4.1. For the pixel-updated road surface image obtained in step S3, select all images containing cracks, peeling, potholes, and bulges; S4.2. By manually labeling the crack area and the defects of spalling, potholes, and lumps in the image, the cracks are marked as Crack, and the other three defects are marked as other1, other2, and other3, respectively, to construct a dataset for real-time detection of pavement cracks. S4.3. The data set for real-time detection of pavement cracks obtained in step S4.2 is divided into a training set and a test set in a ratio of 8:2.

[0030] S5. Construct a fast pavement crack identification method integrating multi-model features; Furthermore, the specific implementation method of step S5 includes the following steps: S5.1. Construct a first deep learning model A; S5.1.1. The first deep learning model A consists of 1 input layer, 4 convolutional layers, 5 activation layers, 3 pooling layers, 1 flatten layer, 1 fully connected layer, and 1 output layer. The specific network structure is the following structure connected in sequence: Input layer: image size 227×227×3; Convolutional layer 1: convolution kernel size 3×3; number of convolution kernels 32; stride: 1; batch normalization; Activation layer 1: The activation function is LeakyReLU; Pooling layer 1: Type is maximum pooling; pooling size: 2×2; Stride: 2; Convolutional layer 2: kernel size 3×3; number of kernels 64; Stride: 1; batch normalization; Activation layer 2: The activation function is LeakyReLU; Pooling layer 2: Type is maximum pooling; pooling size: 2×2; Stride: 2; Convolution layer 3: convolution kernel size 3×3; number of convolution kernels 128; stride: 1; batch normalization; Activation layer 3: The activation function is LeakyReLU; Convolutional layer 4: convolution kernel size 3×3; number of convolution kernels 256; stride: 1; batch normalization; Activation layer 4: The activation function is LeakyReLU; Pooling layer 3: Type is maximum pooling; pooling size: 2×2; Stride: 2; Flatten layer: flattens the output of the last convolutional layer; Fully connected layer: 1024 neurons; Activation layer 5: flattening layer LeakyReLU; Output layer: type is normalized exponential function Softmax; number of neurons: 4; S5.1.2. Input the training set obtained in step S4 into the first deep learning model A, calculate the weights and biases of the neural network units through the error back propagation algorithm, complete the training of the first deep learning model A, and obtain the trained first deep learning model A. The calculation formula of the loss function of the first deep learning model A is: ; in, is the loss function of the first deep learning model A, is the total number of data samples in the first deep learning model A, For the The true labels of samples; The model predicts The probability that a sample is class 1; S5.1.3. The output values ​​of the four output unit nodes corresponding to the first deep learning model A are , , , ; Perform initial conversion on the node output values ​​of the 4 output units to obtain the output value The corresponding probability value The calculation formula is: ; Then the node output values ​​of the four output units are further optimized and converted to obtain the output value The corresponding enhancement probability value The calculation formula is: ; in, G is the enhancement coefficient, and then the actual disease corresponding to each node is marked as , specifically, for road cracks 1, for other diseases is 0; S5.1.4. Adoption And the training set, combined with the error chain derivative rule, train the first deep learning model A. After the training is completed, the output values ​​of the corresponding four types of diseases are obtained, which are recorded as , , , ; S5.2. Constructing a second deep learning model B; S5.2.1. The second deep learning model B consists of 1 input layer, 8 convolutional layers, 9 activation layers, 5 pooling layers, 1 Flatten layer, 1 fully connected layer, and 1 output layer. The specific network structure is the following structure connected in sequence: Input layer: image size 227×227×3; Convolutional layer 1: kernel size 3×3; number of kernels 64; Stride: 1; batch normalization; Activation layer 1: The activation function is LeakyReLU; Pooling layer 1: Type is maximum pooling; pooling size: 2×2; Stride: 2; Convolutional layer 2: convolution kernel size 3×3; number of convolution kernels 128; stride: 1; batch normalization; Activation layer 2: The activation function is LeakyReLU; Pooling layer 2: Type is maximum pooling; pooling size: 2×2; Stride: 2; Convolution layer 3: convolution kernel size 3×3; number of convolution kernels 256; stride: 1; batch normalization; Activation layer 3: The activation function is LeakyReLU; Convolutional layer 4: kernel size 3×3; number of kernels 512; Stride: 1; batch normalization; Activation layer 4: The activation function is LeakyReLU; Pooling layer 3: Type is maximum pooling; pooling size: 2×2; Stride: 2; Convolution layer 5: kernel size 3×3; number of kernels 512; stride: 1; batch normalization; Activation layer 5: The activation function is LeakyReLU; Convolutional layer 6: kernel size 3×3; number of kernels 512; Stride: 1; batch normalization; Activation layer 6: The activation function is LeakyReLU; Pooling layer 4: Type is maximum pooling; pooling size: 2×2; Stride: 2; Convolutional layer 7: kernel size 3×3; number of kernels 1024; stride: 1; batch normalization; Activation layer 7: The activation function is LeakyReLU; Convolutional layer 8: kernel size 3×3; number of kernels 1024; stride: 1; batch normalization; Activation layer 8: The activation function is LeakyReLU; Pooling layer 5: Type is maximum pooling; pooling size: 2×2; Stride: 2; Flatten layer: flattens the output of the last convolutional layer; Fully connected layer: 1024 neurons; Activation layer 9: The activation function is LeakyReLU; Output layer: Type: Softmax; Number of neurons: 4; S5.2.2. Input the training set obtained in step S4 into the second deep learning model B, calculate the weights and biases of the neural network units through the error back propagation algorithm, complete the training of the second deep learning model B, and obtain the trained second deep learning model B. The calculation formula of the loss function of the second deep learning model B is: ; in, is the loss function of the second deep learning model B; is the total number of data samples in the second deep learning model B; For the The true labels of samples, The second deep learning model B predicts The probability that a sample is class 1; S5.2.3. The node output values ​​of the four output units corresponding to the second deep learning model B are b 1. b 2. b 3. b 4. Perform initial conversion on the node output values ​​of the four output units to obtain the output values The corresponding probability value The calculation formula is: ; Then the node output values ​​of the four output units are further optimized and converted to obtain the output value b i The corresponding enhancement probability value The calculation formula is: ; S5.2.4. Adoption The training set obtained in step S4 is combined with the error chain derivation rule to train the second deep learning model B. After the training is completed, the disease output value corresponding to the second deep learning model B is obtained as follows: , among the 4 nodes in the output layer, the actual disease corresponding to each node is marked as , =1,2,3,4, specifically, for road cracks 1, for other diseases is 0; S5.3. Constructing a third deep learning model C; S5.3.1. The third deep learning model C consists of 1 input layer, 12 convolutional layers, 13 activation layers, 6 pooling layers, 1 Flatten layer, 1 fully connected layer, and 1 output layer. The specific network structure is the following structure connected in sequence: Input layer: image size 227×227×3; Convolutional layer 1: kernel size 3×3; number of kernels 64; Stride: 1; batch normalization; Activation layer 1: The activation function is LeakyReLU; Convolutional layer 2: kernel size 3×3; number of kernels 64; Stride: 1; batch normalization; Activation layer 2: The activation function is LeakyReLU; Pooling layer 1: Type is maximum pooling; pooling size: 2×2; Stride: 2; Convolution layer 3: convolution kernel size 3×3; number of convolution kernels 128; stride: 1; batch normalization; Activation layer 3: The activation function is LeakyReLU; Convolutional layer 4: convolution kernel size 3×3; number of convolution kernels 128; stride: 1; batch normalization; Activation layer 4: The activation function is LeakyReLU; Pooling layer 2: Type is maximum pooling; pooling size: 2×2; Stride: 2; Convolution layer 5: kernel size 3×3; number of kernels 256; stride: 1; batch normalization; Activation layer 5: The activation function is LeakyReLU; Convolutional layer 6: kernel size 3×3; number of kernels 256; stride: 1; batch normalization; Activation layer 6: The activation function is LeakyReLU; Pooling layer 3: Type is maximum pooling; pooling size: 2×2; Stride: 2; Convolutional layer 7: kernel size 3×3; number of kernels 512; stride: 1; batch normalization; Activation layer 7: The activation function is LeakyReLU; Convolutional layer 8: kernel size 3×3; number of kernels 512; Stride: 1; batch normalization; Activation layer 8: The activation function is LeakyReLU; Pooling layer 4: Type is maximum pooling; pooling size: 2×2; Stride: 2; Convolutional layer 9: kernel size 3×3; number of kernels 1024; stride: 1; batch normalization; Activation layer 9: The activation function is LeakyReLU; Convolution layer 10: kernel size 3×3; number of kernels 1024; stride: 1; batch normalization; Activation layer 10: The activation function is LeakyReLU; Pooling layer 5: Type is maximum pooling; pooling size: 2×2; Stride: 2; Convolution layer 11: kernel size 3×3; number of kernels 2048; stride: 1; batch normalization; Activation layer 11: The activation function is LeakyReLU; Convolutional layer 12: kernel size 3×3; number of kernels 2048; stride: 1; batch normalization; Activation layer 12: The activation function is LeakyReLU; Pooling layer 6: Type is maximum pooling; pooling size: 2×2; Stride: 2; Flatten layer: flattens the output of the last convolutional layer; Fully connected layer: 1024 neurons; Activation layer 13: The activation function is LeakyReLU; Output layer: type is Softmax; number of neurons: 4; S5.3.2. Input the training set obtained in step S4 into the third deep learning model C, calculate the weights and biases of the neural network units through the error back propagation algorithm, complete the training of the third deep learning model C, and obtain the trained third deep learning model C; The calculation formula of the loss function of the third deep learning model C is: ; in, is the loss function of the third deep learning model C; is the total number of data samples in the third deep learning model C; For the The true labels of samples, The third deep learning model C predicts The probability that a sample is class 1; S5.3.3. The node output values ​​of the four output units corresponding to the third deep learning model C are , the node output values ​​of the four output units are initially converted to obtain The corresponding probability value The calculation formula is: ; Then the node output values ​​of the four output units are further optimized and converted to obtain the output value The corresponding enhancement probability value The calculation formula is: ; in, G is the enhancement coefficient, and then the actual disease corresponding to each node is marked as , specifically, for road cracks 1, for other diseases is 0; S5.3.4. Adoption The training set obtained in step S4 is combined with the error chain derivation rule to train the third deep learning model C. After the training is completed, the output values ​​of the four types of diseases corresponding to the third deep learning model C are obtained as follows: .

[0031] S5.4. Based on and , establish the loss function between the second deep learning model B and the output layer of the first deep learning model A, and the calculation formula is: ; based on and , construct the output value The corresponding probability value The loss function of the first deep learning model A , the calculation formula is: ; based on and , establish the relationship loss function between the second deep learning model B and the output layer of the first deep learning model A , the calculation formula is: , in, for The weight coefficient of for The weight coefficient of S5.5. Establish the relationship loss function between the output layer of the third deep learning model C and the second deep learning model B , the expression is: ; in, for The weight coefficient of for The weight coefficient of Based on and The loss function of Based on and The loss function of The calculation formula is: ; The calculation formula is: ; S5.6. Establish the relationship loss function between the output layer of the third deep learning model C and the first deep learning model A , the expression is: ; in, for The weight coefficient of for The weight coefficient of Based on and The loss function is Based on and The loss function of The calculation formula is: ; The calculation formula is: ; S5.7. Define the overall loss function corresponding to the first deep learning model A, the second deep learning model B, and the third deep learning model C , the calculation formula is: ; S5.8. Using the data set for rutting recognition obtained in step S4, based on the chain rule and through error back propagation, the first deep learning model A, the second deep learning model B, and the third deep learning model C are trained. When the overall loss function obtained in step S5.7 is When is the minimum, the corresponding model parameters are optimal, and the first deep learning A after training is the pavement crack rapid identification model with comprehensive multi-model features that is finally used.

[0032] The integration of digital twin technology and rapid pavement crack identification methods can use high-precision digital models combined with real-time data to achieve instant and accurate diagnosis of pavement conditions. By integrating sensor data, image recognition technology, and machine learning models, it can not only detect and locate pavement cracks in a timely manner, but also predict future damage development, thereby optimizing maintenance plans and resource allocation. This helps improve road safety, reduce maintenance costs, and extend the service life of roads, which is of great value in improving public safety and economic efficiency.

[0033] S6. Constructing a road digital twin model based on a road crack rapid identification method integrating multi-model features in step S5; Furthermore, the specific implementation method of step S6 includes the following steps: S6.1. Export the pavement crack rapid identification model with integrated multi-model features obtained in step S5 as a deployment file for independent services, including one of a pt file, a pth file, a json file, and a txt file; S6.2. Construct a road information model; S6.2.1. Collect information on the location, length, width, and layout of roads; S6.2.2. The data obtained in step S6.2.1 is stored in a database, and a basic three-dimensional model of the road is constructed, wherein the basic three-dimensional model of the road includes geometric information of the road; S6.2.3. Based on step S6.2.2, material properties, design specifications, and historical maintenance data are added to the basic three-dimensional model of the road to obtain a road information model; S6.2.4. Export the road information model to a format file supported by the web application, including a json file or a js file; S6.3. In the digital twin platform, import the deployment file of the pavement crack rapid identification model with integrated multi-model features obtained in step S6.1, and import the road information model format file obtained in step S7.2; S6.4. Use Python to construct an API interface between image data and a road crack rapid identification model with integrated multi-model features for intelligent rutting identification; Use Python to construct an API interface between a road crack rapid identification model with integrated multi-model features and a road information model to complete the construction of a road digital twin model; S7. Based on the road digital twin model obtained in step S6, a pavement crack classification method is established.

[0034] Furthermore, the specific implementation method of step S7 includes the following steps: S7.1. Use a high-definition camera on a vehicle to collect road images and upload them to the road digital twin model constructed in step S6, intelligently identify crack areas in the road images, and select all images containing road cracks; S7.2. For all pavement crack regions of the image of pavement cracks identified in step S7.1, obtain the length of the cracks by on-site verification and measurement. CD , Width data KD ; Then, collect crack information at different times, compare crack length and width data, and calculate length development speed CF , Width Development Speed KF ; S7.3. Construct a crack criteria matrix based on the length of the cracks CD ,width KD , length development speed CF , Width Development Speed KF ,quantity SL , establish the road pavement crack state classification standard, divide the crack state into NL categories, and obtain the length standard value, width standard value, length development speed standard value, width development speed standard value, and quantity standard value, which are recorded as S7.3. Construct the crack standard matrix based on the length of the crack CD ,width KD , length development speed CF , Width Development Speed KF ,quantity SL , a road pavement crack state classification standard is established, and the crack state is divided into NL categories. The length standard value, width standard value, length growth speed standard value, width growth speed standard value, and quantity standard value are recorded as , ; Establish a crack standard matrix based on the standard value of each type of crack, Crack-like standard matrix The expression is: ; S7.3. Classify pavement cracks; S7.3.1. Collect road surface images for the NW road, use the road digital twin model to identify images containing road surface cracks, and obtain the first l The size information of the road cracks on the road, including length, width, and number, is collected and calculated at different times to calculate the length and width growth rates of the cracks. On this basis, the first l Crack matrix of the road , the expression is: ; S7.3.2. Calculate the l The crack matrix of the first road is Crack Criteria Matrix The distance between , the calculation formula is: ; in, is the weight coefficient between the crack length and the standard value of length, is the weight coefficient between the crack width and the standard value of the width, is the weight coefficient between the crack length development speed and the standard value of length development speed, is the weight coefficient between the crack width development speed and the standard value of the width development speed, is the weight coefficient between the number of cracks and the standard value of the number; S7.3.3. Calculate the l The distance between the crack matrix of the road and all the crack standard matrices is calculated, and the maximum value is found. The crack standard corresponding to the maximum value is l Classification of road pavement crack states corresponding to the roads; S7.3.4. Based on the methods of steps S7.3.2 and S7.3.3, the distances between all crack standard matrices corresponding to all NW roads are calculated to complete the classification of road pavement crack states of all roads.

[0035] The key points and intended protection points of the present invention are: (1) A method for updating the pixels of pavement crack images; (2) A rapid pavement crack identification method that integrates multi-model features.

[0036] (3) A pavement crack classification method based on digital twins.

[0037] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0038] Although the present application has been described above with reference to specific embodiments, various modifications may be made thereto and parts thereof may be replaced with equivalents without departing from the scope of the present application. In particular, as long as there is no structural conflict, the various features in the specific embodiments disclosed in the present application may be used in combination with each other in any manner, and the fact that these combinations are not exhaustively described in this specification is only for the sake of omitting space and saving resources. Therefore, the present application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A real-time detection and classification method for pavement cracks using digital twin technology, characterized in that: The steps include: S1. Use a high-definition camera on a vehicle to collect road images and save them as an image matrix; S2. Perform image quality enhancement processing on the image matrix acquired in step S1 to obtain a quality-enhanced image matrix; S3. Performing image pixel updates on the quality enhanced image matrix obtained in step S2 to obtain a pixel updated image matrix, which is saved as a pixel updated road surface image; S4. For the pixel updated road image obtained in step S3, all images including cracks, peeling, potholes, and bagging diseases are selected and annotated to obtain a data set for real-time detection of road cracks; S5. Construct a fast pavement crack identification method integrating multi-model features; S6. Constructing a road digital twin model based on a road crack rapid identification method integrating multi-model features in step S5; S7. Based on the road digital twin model obtained in step S6, a pavement crack classification method is established.

2. The real-time detection and classification method for pavement cracks using digital twin technology according to claim 1 is characterized in that: The road surface image collected in step S1 is denoted as ph, and the corresponding image matrix is ​​denoted as PH .

3. The real-time detection and classification method for pavement cracks using digital twin technology according to claim 2 is characterized in that: The specific implementation method of step S2 includes the following steps: S2.

1. Constructing the 5th-order denoising kernel function matrix N , the calculation formula is: ； in, is the element in the kernel function matrix, is the row number of the 5th-order denoising kernel function matrix, is the column number of the 5th-order denoising kernel function matrix; ； in, is the standard deviation of all elements in the kernel function; S2.

2. Normalize all elements in the kernel function matrix to obtain the normalized denoising kernel function matrix , the calculation formula is: ； S2.

3. Perform denoising kernel function processing on the image matrix collected in step S1 to obtain the image matrix processed by denoising kernel function P , the calculation formula is: 。 4. The real-time detection and classification method for pavement cracks using digital twin technology according to claim 3 is characterized in that: The specific implementation method of step S3 includes the following steps: S3.

1. Divide the pixel grayscale value 0~255 into 16 grayscale intervals, denoted as , is the gray level number, =1,2,3,…,16; S3.

2. The image matrix processed by the denoising kernel function obtained in step S2 is divided into N equal parts. The image matrix after equal division is recorded as P num, num is the number of the image matrix after equal division, num=1,2,…,N; S3.

3. Count each gray level interval in the equally divided image matrix The frequency of the corresponding pixel ; S3.

4. Calculate each gray level in Pnum The corresponding grayscale frequency cumulative value , the calculation formula is: ； in, is less than or equal to The gray level number, =0,1,2,…, ; S3.

5. Construct a pixel value update method, the calculation formula is: ； in, is the minimum non-zero value among all frequency accumulation values, is the image matrix after equal division P The total number of pixels corresponding to num, is the maximum value of the gray level, Grayscale The corresponding pixel update value; S3.

6. Perform pixel value update processing on all equally divided image matrices to obtain pixel-updated image matrices, and save them as pixel-updated road surface images.

5. The real-time detection and classification method for pavement cracks using digital twin technology according to claim 4 is characterized in that: The specific implementation method of step S4 includes the following steps: S4.

1. For the pixel-updated road surface image obtained in step S3, select all images containing cracks, peeling, potholes, and bulges; S4.

2. By manually labeling the crack area and the defects of spalling, potholes, and lumps in the image, the cracks are marked as Crack, and the other three defects are marked as other1, other2, and other3, respectively, to construct a dataset for real-time detection of pavement cracks. S4.

3. The data set for real-time detection of pavement cracks obtained in step S4.2 is divided into a training set and a test set in a ratio of 8:

2.

6. The real-time detection and classification method for pavement cracks using digital twin technology according to claim 5 is characterized in that: The specific implementation method of step S5 includes the following steps: S5.

1. Construct a first deep learning model A; S5.1.

1. The first deep learning model A consists of 1 input layer, 4 convolutional layers, 5 activation layers, 3 pooling layers, 1 flatten layer, 1 fully connected layer, and 1 output layer; S5.1.

2. Input the training set obtained in step S4 into the first deep learning model A, calculate the weights and biases of the neural network units through the error back propagation algorithm, complete the training of the first deep learning model A, and obtain the trained first deep learning model A. The calculation formula of the loss function of the first deep learning model A is: ； in, is the loss function of the first deep learning model A, is the total number of data samples in the first deep learning model A, For the The true labels of samples; The model predicts The probability that a sample is class 1; S5.1.

3. The output values ​​of the four output unit nodes corresponding to the first deep learning model A are , , , ; Perform initial conversion on the node output values ​​of the 4 output units to obtain the output value The corresponding probability value The calculation formula is: ； Then the node output values ​​of the four output units are further optimized and converted to obtain the output value The corresponding enhancement probability value The calculation formula is: ； in, G is the enhancement coefficient, and then the actual disease corresponding to each node is marked as ; S5.1.

4. Adoption And the training set, combined with the error chain derivative rule, train the first deep learning model A. After the training is completed, the output values ​​of the corresponding four types of diseases are obtained, which are recorded as , , , ; S5.

2. Constructing a second deep learning model B; S5.2.

1. The second deep learning model B consists of 1 input layer, 8 convolutional layers, 9 activation layers, 5 pooling layers, 1 Flatten layer, 1 fully connected layer, and 1 output layer; S5.2.

2. Input the training set obtained in step S4 into the second deep learning model B, calculate the weights and biases of the neural network units through the error back propagation algorithm, complete the training of the second deep learning model B, and obtain the trained second deep learning model B. The calculation formula of the loss function of the second deep learning model B is: ； in, is the loss function of the second deep learning model B; is the total number of data samples in the second deep learning model B; For the The true labels of samples, The second deep learning model B predicts The probability that a sample is class 1; S5.2.

3. The node output values ​​of the four output units corresponding to the second deep learning model B are b 1. b 2. b 3. b 4. Perform initial conversion on the node output values ​​of the four output units to obtain the output values The corresponding probability value The calculation formula is: ； Then the node output values ​​of the four output units are further optimized and converted to obtain the output value b i The corresponding enhancement probability value The calculation formula is: ； S5.2.

4. Adoption The training set obtained in step S4 is combined with the error chain derivation rule to train the second deep learning model B. After the training is completed, the disease output value corresponding to the second deep learning model B is obtained as follows: , among the 4 nodes in the output layer, the actual disease corresponding to each node is marked as , =1,2,3,4; S5.

3. Constructing a third deep learning model C; S5.3.

1. The third deep learning model C consists of 1 input layer, 12 convolutional layers, 13 activation layers, 6 pooling layers, 1 Flatten layer, 1 fully connected layer, and 1 output layer; S5.3.

2. Input the training set obtained in step S4 into the third deep learning model C, calculate the weights and biases of the neural network units through the error back propagation algorithm, complete the training of the third deep learning model C, and obtain the trained third deep learning model C; The calculation formula of the loss function of the third deep learning model C is: ； in, is the loss function of the third deep learning model C; is the total number of data samples in the third deep learning model C; For the The true labels of samples, The third deep learning model C predicts The probability that a sample is class 1; S5.3.

3. The node output values ​​of the four output units corresponding to the third deep learning model C are , the node output values ​​of the four output units are initially converted to obtain The corresponding probability value The calculation formula is: ； Then the node output values ​​of the four output units are further optimized and converted to obtain the output value The corresponding enhancement probability value The calculation formula is: ； in, G is the enhancement coefficient, and then the actual disease corresponding to each node is marked as ; S5.3.

4. Adoption The training set obtained in step S4 is combined with the error chain derivation rule to train the third deep learning model C. After the training is completed, the output values ​​of the four types of diseases corresponding to the third deep learning model C are obtained as follows: ; S5.

4. Based on and , establish the loss function between the second deep learning model B and the output layer of the first deep learning model A, and the calculation formula is: ； based on and , construct the output value The corresponding probability value The loss function of the first deep learning model A , the calculation formula is: ； based on and , establish the relationship loss function between the second deep learning model B and the output layer of the first deep learning model A , the calculation formula is: ， in, for The weight coefficient of for The weight coefficient of S5.

5. Establish the relationship loss function between the output layer of the third deep learning model C and the second deep learning model B , the expression is: ； in, for The weight coefficient of for The weight coefficient of Based on and The loss function of Based on and The loss function of The calculation formula is: ； The calculation formula is: ； S5.

6. Establish the relationship loss function between the output layer of the third deep learning model C and the first deep learning model A , the expression is: ； in, for The weight coefficient of for The weight coefficient of Based on and The loss function is Based on and The loss function of The calculation formula is: ； The calculation formula is: ； S5.

7. Define the overall loss function corresponding to the first deep learning model A, the second deep learning model B, and the third deep learning model C , the calculation formula is: ； S5.

8. Using the data set for rutting recognition obtained in step S4, based on the chain rule and through error back propagation, the first deep learning model A, the second deep learning model B, and the third deep learning model C are trained. When the overall loss function obtained in step S5.7 is When is the minimum, the corresponding model parameters are optimal, and the first deep learning A after training is the pavement crack rapid identification model with comprehensive multi-model features that is finally used.

7. The real-time detection and classification method for pavement cracks using digital twin technology according to claim 6 is characterized in that: The specific implementation method of step S6 includes the following steps: S6.

1. Export the pavement crack rapid identification model with integrated multi-model features obtained in step S5 as a deployment file for independent services, including one of a pt file, a pth file, a json file, and a txt file; S6.

2. Construct a road information model; S6.2.

1. Collect information on the location, length, width, and layout of roads; S6.2.

2. The data obtained in step S6.2.1 is stored in a database, and a basic three-dimensional model of the road is constructed, wherein the basic three-dimensional model of the road includes geometric information of the road; S6.2.

3. Based on step S6.2.2, material properties, design specifications, and historical maintenance data are added to the basic three-dimensional model of the road to obtain a road information model; S6.2.

4. Export the road information model to a format file supported by the web application, including a json file or a js file; S6.

3. In the digital twin platform, import the deployment file of the pavement crack rapid identification model with integrated multi-model features obtained in step S6.1, and import the road information model format file obtained in step S7.2; S6.

4. Use Python to construct an API interface between image data and a road surface crack rapid identification model with integrated multi-model features for intelligent rutting identification. Use Python to construct an API interface between a road surface crack rapid identification model with integrated multi-model features and a road information model to complete the construction of a road digital twin model.

8. The real-time detection and classification method for pavement cracks using digital twin technology according to claim 7 is characterized in that: The specific implementation method of step S7 includes the following steps: S7.

1. Use a high-definition camera on a vehicle to collect road images and upload them to the road digital twin model constructed in step S6, intelligently identify crack areas in the road images, and select all images containing road cracks; S7.

2. For all pavement crack regions of the image of pavement cracks identified in step S7.1, obtain the length of the cracks by on-site verification and measurement. CD , Width data KD ; Then, collect crack information at different times, compare crack length and width data, and calculate length development speed CF , Width Development Speed KF ; S7.

3. Construct a crack criteria matrix based on the length of the cracks CD ,width KD , length development speed CF , Width Development Speed KF ,quantity SL , a road pavement crack state classification standard is established, and the crack state is divided into NL categories. The length standard value, width standard value, length growth speed standard value, width growth speed standard value, and quantity standard value are recorded as , ; A crack standard matrix is ​​established based on the standard value of each type of crack. Crack-like standard matrix The expression is: ； S7.

3. Classify pavement cracks; S7.3.

1. Collect road surface images for the NW road, use the road digital twin model to identify images containing road surface cracks, and obtain the first l The size information of the road cracks on the road, including length, width, and number, is collected and calculated at different times to calculate the length and width growth rates of the cracks. On this basis, the first l Crack matrix of the road , the expression is: ； S7.3.

2. Calculate the l The crack matrix of the first road is Crack Criteria Matrix The distance between , the calculation formula is: ； in, is the weight coefficient between the crack length and the standard value of length, is the weight coefficient between the crack width and the standard value of the width, is the weight coefficient between the crack length development speed and the standard value of length development speed, is the weight coefficient between the crack width development speed and the standard value of the width development speed, is the weight coefficient between the number of cracks and the standard value of the number; S7.3.

3. Calculate the l The distance between the crack matrix of the road and all the crack standard matrices is calculated, and the maximum value is found. The crack standard corresponding to the maximum value is l Classification of road pavement crack states corresponding to the roads; S7.3.

4. Based on the methods of steps S7.3.2 and S7.3.3, the distances between all crack standard matrices corresponding to all NW roads are calculated to complete the classification of road pavement crack states of all roads.

Images