Pavement distress detection method and apparatus based on yolov9 model improvement

Abstract

Disclosed in the present invention are a pavement distress detection method and apparatus based on YOLOv9 model improvement. The pavement distress detection method comprises: inputting a pavement distress image to be inspected into a trained pavement distress model based on YOLOv9 improvement, and performing identification, so as to obtain a detection result. A method for training the pavement distress model based on YOLOv9 improvement comprises: acquiring a pavement distress image data set, and dividing same into a training set and a verification set; adding an LSKNet module after a specific layer of a YOLOv9 model, and constructing a pavement distress detection model based on YOLOv9 improvement; and using the training set and the verification set to train the constructed pavement distress detection model, so as to obtain the trained pavement distress model based on YOLOv9 improvement. In the present invention, a header network and a backbone network of YOLOv9 are improved, such that YOLOv9 can capture more extensive context information from an image, thereby improving the detection accuracy of pavement distress.

Description

A method and apparatus for detecting pavement distress based on the improved YOLOv9 model Technical Field

[0001] This invention relates to a method and apparatus for detecting pavement defects based on an improved YOLOv9 model, belonging to the field of information perception and recognition technology. Background Technology

[0002] Vision-based road defect detection can collect data from roads using devices such as dashcams, vehicle-mounted cameras, and drones, offering advantages such as low cost, high speed, and high accuracy, and is currently widely used. However, in actual detection work, due to complex road backgrounds and the influence of lighting conditions on imaging, missed detections and false detections frequently occur. Therefore, improving the accuracy of road defect recognition is an important issue in the field of computer vision.

[0003] Target detection networks used for road surface defect identification can be broadly categorized into two types: one-stage target detection networks and two-stage target detection networks. One-stage target detection networks offer advantages over two-stage networks in terms of speed and real-time performance, making them suitable for industrial inspection applications with high real-time requirements. In the realm of one-stage target detection networks, the YOLO series has consistently garnered significant attention. Since the release of its first generation, the YOLO model has undergone numerous updates, each boasting improved detection efficiency and accuracy. The latest version, YOLOv9, has achieved a major breakthrough in both speed and accuracy, becoming the state-of-the-art (SOTA) model in the field of target detection. However, the YOLOv9 model still faces the challenge of focusing effectively on key regions during learning. To address this, this invention introduces the LSKNet network, which integrates a large kernel selection module (LK Selection) and a feedforward network module (FFN). The LK Selection module dynamically adjusts the network's receptive field as needed, while the FFN module is used for channel blending and feature refinement. The introduction of the LSKNet network effectively solves the problem that the model cannot focus well on key areas, thus improving the detection accuracy of road surface defects.

[0004] Currently, there is still a lack of research on how to improve the accuracy of pavement distress detection based on the YOLOv9 model.

[0005] Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a pavement distress detection method and device based on the improved YOLOv9 model, so as to achieve higher accuracy in detecting pavement distress.

[0007] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0008] In a first aspect, the present invention provides a pavement distress detection method based on an improved YOLOv9 model, comprising:

[0009] The images of road surface defects to be detected are input into the trained road surface defect model based on YOLOv9 for identification, and the detection results are obtained.

[0010] The training method for the pavement distress model based on the improved YOLOv9 includes:

[0011] Obtain a dataset of road surface damage images and divide it into a training set and a validation set;

[0012] An LSKNet module was added after a specific layer of the YOLOv9 model to construct a pavement distress detection model based on an improved YOLOv9.

[0013] The constructed pavement distress detection model was trained using the training and validation sets to obtain the trained pavement distress model based on YOLOv9.

[0014] Furthermore, the step of adding an LSKNet module after a specific layer of the YOLOv9 model to construct an improved pavement distress detection model based on YOLOv9 includes:

[0015] An LSKNet module is constructed, which includes a large kernel selection module (LK Selection) and a feedforward network module (FFN). The LK Selection module can dynamically adjust the receptive field of the network as needed, and the FFN module is used for channel blending and feature refinement.

[0016] The LSKNet module is added after the RepNCSPELAN4 module at layers 16, 19, and 22 of the YOLOv9 model's head network, and the LSKNet module is added before the SPPELAN module at layer 11 of the YOLOv9 model's backbone network, resulting in a pavement distress detection model improved based on YOLOv9.

[0017] Furthermore, the LK Selection module includes a fully connected layer sequence, a GELU activation function layer, a core LSK layer, and a second fully connected layer.

[0018] Furthermore, the FFN module includes a sequence of fully connected layers, a depthwise convolution, a GELU activation function layer, and a second fully connected layer.

[0019] Furthermore, the road surface defect image dataset is derived from the RDD2020 competition dataset.

[0020] Furthermore, the pavement distress image dataset contains multiple categories of pavement distress, namely Longitudinal Cracks (D00), Transverse Cracks (D10), Alligator Cracks (D20), and Potholes (D40).

[0021] Furthermore, the detection results include information on the type and location of road surface defects.

[0022] Furthermore, the method also includes: dividing a test set from the road surface defect image dataset, and using the test set to evaluate the detection accuracy of the trained road surface defect detection model based on YOLOv9.

[0023] Secondly, the present invention provides a pavement distress detection device based on an improved YOLOv9 model, comprising:

[0024] The detection module is used to input the pavement distress image to be detected into the trained pavement distress model based on YOLOv9 for identification and to obtain the detection results.

[0025] The training method for the pavement distress model based on YOLOv9, as described in the detection module, includes:

[0026] Obtain a dataset of road surface damage images and divide it into a training set and a validation set;

[0027] An LSKNet module was added after a specific layer of the YOLOv9 model to construct a pavement distress detection model based on an improved YOLOv9.

[0028] The constructed pavement distress detection model was trained using the training and validation sets to obtain the trained pavement distress model based on YOLOv9.

[0029] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0030] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0031] (1) The detection method proposed in this invention is based on the state-of-the-art YOLOv9 target detection algorithm, and introduces the LSKNet module to make it have higher detection accuracy. In this invention, the LSKNet module is inserted into the head network and backbone network of YOLOv9, enabling the model to capture a wider range of contextual information in the image and improve the detection accuracy of road surface defects.

[0032] (2) The present invention proposes a large kernel selection module LK Selection to enhance feature expression capabilities. This module dynamically adjusts the receptive field of the network as needed to better capture and identify irregular diseases.

[0033] (3) The present invention proposes an FFN feedforward network module for feature refinement, which can learn higher-level feature representations, thereby improving the expressive power of the model. At the same time, the module also introduces position awareness by independently transforming the features at each position, so that the model can perform weighted processing on the features at different positions according to the position information, thereby enhancing the model's attention to different positions in the image.

[0034] In summary, this invention employs a fusion of multiple advanced modules, resulting in a network architecture that better meets the requirements of pavement distress detection. By improving the head and backbone networks of YOLOv9, it is able to capture a wider range of contextual information in images, thereby enhancing the accuracy of pavement distress detection. Attached Figure Description

[0035] Figure 1 is a flowchart of the present invention;

[0036] Figure 2 is a schematic diagram of the overall network structure of the present invention;

[0037] Figure 3 is a schematic diagram of the LSKNet core module in this invention;

[0038] Figure 4 is a schematic diagram of the LSK selectivity mechanism module in this invention;

[0039] Figure 5 is a schematic diagram of road surface defects in this invention;

[0040] Figure 6 is a schematic diagram of the pavement distress detection results in this invention. Detailed Implementation

[0041] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0042] Example 1

[0043] This embodiment introduces a pavement distress detection method based on an improved YOLOv9 model, including:

[0044] The images of road surface defects to be detected are input into the trained road surface defect model based on YOLOv9 for identification, and the detection results are obtained.

[0045] The training method for the pavement distress model based on the improved YOLOv9 includes:

[0046] Obtain a dataset of road surface damage images and divide it into a training set and a validation set;

[0047] An LSKNet module was added after a specific layer of the YOLOv9 model to construct a pavement distress detection model based on an improved YOLOv9.

[0048] The constructed pavement distress detection model was trained using the training and validation sets to obtain the trained pavement distress model based on YOLOv9.

[0049] The pavement distress detection method based on the improved YOLOv9 model provided in this embodiment involves the following steps in its application:

[0050] Step 1: Obtain the pavement distress image dataset and divide it into training, validation, and test sets. This invention uses the RDD2020 competition dataset, which contains multiple categories of pavement distress, but this invention focuses only on four types: Longitudinal Cracks (D00), Transverse Cracks (D10), Alligator Cracks (D20), and Potholes (D40). The dataset is located at: https: / / data.mendeley.com / datasets / 5ty2wb6gvg / 1.

[0051] Step 2: Construct the LSKNet module. Referring to the schematic diagram of the LSKNet core module shown in Figure 3, the LSKNet module includes a large kernel selection module (LK Selection) and a feedforward network module (FFN).

[0052] The large kernel selection module (LK Selection) dynamically adjusts the receptive field of the network as needed. It consists of a fully connected layer sequence (FC), a GELU activation function layer, a core LSK layer, and a second fully connected layer. The input of the first FC module is connected to external input, and the output of the fourth FC module is added to the input of the layer above the LK Selection sub-block as its output.

[0053] The feedforward network module FFN, used for channel blending and feature refinement, consists of a fully connected layer sequence (FC), a depthwise convolutional layer, a GELU activation function layer, and a second fully connected layer. The input of the first FC module is connected to an external input, and the output of the fourth FC module is added to the input of the layer above the FFN sub-block as its output.

[0054] Step 3: Referring to the overall network structure diagram shown in Figure 2, construct a pavement distress detection model based on YOLOv9. Specifically, for the head network of the YOLOv9 model, add LSKNet modules after the RepNCSPELAN4 modules at layers 16, 19, and 22, respectively. For the backbone network of the YOLOv9 model, add an LSKNet module before the SPPELAN module at layer 11, thus obtaining a pavement distress detection model based on YOLOv9.

[0055] Referring to the schematic diagram of the LSK selectivity mechanism module shown in Figure 4, adding the LSK attention mechanism to the residual network as described above can help the network better learn and utilize key information in the input data, thereby enhancing the expressive power of the features. At the same time, it can also improve the network's flexibility and adaptability, allowing the feature weights to be dynamically adjusted according to the characteristics of the input, making the network more flexible in adapting to different input data. Adding the LSK attention mechanism before the feature pyramid network can help the model better fuse features at different scales.

[0056] Step 4: Train the pavement distress detection model built in Step 3. Put the training set and validation set from the dataset in Step 1 into the pavement distress detection model based on YOLOv9 for training to obtain the trained model.

[0057] Step 5: Evaluate the model. Based on the YOLOv9-based improved pavement distress detection model obtained after training, evaluate the detection accuracy of the model.

[0058] Step 6: Input the pavement distress image to be detected into the trained pavement distress model based on YOLOv9 for identification and obtain the detection results. Refer to Figure 6 for a schematic diagram of the pavement distress detection results.

[0059] The following description, in conjunction with a preferred embodiment, illustrates the content involved in the above embodiments.

[0060] The application scenario of this invention is as follows: Most highways in my country are asphalt pavements. Due to their large voids, poor stability, and excessive traffic load, these pavements are prone to cracks, depressions, and other defects. During road maintenance, it is necessary to detect these defects in order to repair and maintain them. The main content of this invention is the research on a pavement defect detection algorithm based on the YOLOv9 model, and the optimization of existing detection networks to achieve higher accuracy in defect detection.

[0061] As shown in Figure 1, this figure clearly illustrates the flowchart of a pavement distress detection method based on YOLOv9 improvement provided by the present invention. According to the flowchart, the method includes the following steps:

[0062] Step 1: Obtain the pavement damage image dataset and divide it into training set, validation set and test set;

[0063] In step 1, the obtained pavement distress image dataset can be a publicly available dataset or a dataset that has been collected and annotated independently. The dataset should contain pavement distress images with annotation information for use in the subsequent training phase of the improved YOLOv9 model. The data used in this invention comes from the RDD2020 competition dataset.

[0064] When using a public dataset, data preprocessing should be performed, including the following steps:

[0065] Step 1.1: The RDD2020 competition dataset used in this invention contains multiple categories of pavement defects, but this invention only focuses on four types: Longitudinal Cracks (D00), Transverse Cracks (D10), Alligator Cracks (D20), and Potholes (D40). The data was cleaned to remove unlabeled images and images not containing the above four categories, effectively improving the model's generalization ability and robustness.

[0066] Step 1.2: Divide the dataset into training, validation, and test sets in an 8:1:1 ratio. The training set will be used as training data for the subsequently improved YOLOv9 model.

[0067] Step 2: Construct the LSKNet module. Referring to the schematic diagram of the LSKNet module structure shown in Figure 3, this module includes a large kernel selection module (LK Selection) and a feedforward network module (FFN);

[0068] The large kernel selection module (LK Selection) dynamically adjusts the receptive field of the network as needed. It consists of a fully connected layer sequence (FC), a GELU activation function layer, a core LSK layer, and a second fully connected layer. The input of the first FC module is connected to external input, and the output of the fourth FC module is added to the input of the layer above the LK Selection sub-block as its output.

[0069] This module constructs larger kernel convolutions by explicitly decomposing them into depthwise convolution sequences with large-growing kernels and increased dilation. Specifically, for the i-th depthwise convolution, with kernel size k and dilation rate d, the receptive field RF is expressed as shown in equations (1) and (2): k i-1 ≤k i ;d1=1,d i-1 <d i ≤RFi-1 (1) RF1=k1,RF i =d i (k i -1)+RF i-1 (2)

[0070] The increased kernel size and dilation rate ensure that the receptive field expands quickly enough. Furthermore, the LSKNet module sets an upper limit on the dilation rate to prevent the dilated convolutions from introducing gaps between feature maps. For example, a large kernel can be decomposed into 2 or 3 depthwise convolutions. This proposed design has two advantages. First, it explicitly generates multiple features with different large receptive fields, making subsequent kernel selection easier. Second, sequential decomposition is more efficient than simply applying a single, larger kernel. Under the same theoretical receptive field, such decomposition significantly reduces the number of parameters compared to standard large convolutional kernels.

[0071] To obtain features with rich contextual information within different ranges of the input X, a series of decomposition depthwise convolutions with different receptive fields were applied:

[0072] in, It has a kernel k i and expansion d i Depthwise convolution. Assume there are N decomposition kernels, each consisting of a 1×1 convolutional layer. Further processing:

[0073] Channel mixing is allowed for each spatial feature vector.

[0074] The feedforward network module FFN, used for channel blending and feature refinement, consists of a fully connected layer sequence (FC), a depthwise convolutional layer, a GELU activation function layer, and a second fully connected layer. The input of the first FC module is connected to an external input, and the output of the fourth FC module is added to the input of the layer above the FFN sub-block as its output.

[0075] Step 3: Construct a pavement distress detection model based on YOLOv9. Specifically, for the head network of the YOLOv9 model, add LSKNet modules after the RepNCSPELAN4 modules at layers 16, 19, and 22, respectively. For the backbone network of the YOLOv9 model, add an LSKNet module before the SPPELAN module at layer 11, thus obtaining a pavement distress detection model based on YOLOv9.

[0076] Adding an LSK attention mechanism to the residual network as described above can help the network better learn and utilize key information in the input data, thereby enhancing the expressive power of the features. At the same time, it can also improve the network's flexibility and adaptability, allowing it to dynamically adjust feature weights based on input characteristics, making the network more flexible in adapting to different input data. Adding an LSK attention mechanism before the feature pyramid network can help the model better fuse features at different scales.

[0077] Referring to the schematic diagram of the LSK selectivity mechanism module in Figure 4, this mechanism explicitly relies on a series of large kernels, which differs from most existing attention-based methods. Secondly, this method adaptively aggregates information from large kernels in the spatial dimension, rather than utilizing the channel dimension. This design is more intuitive and effective because channel selection cannot model the variance of different targets in the image space.

[0078] Step 4: Train the pavement distress detection model based on YOLOv9 built in Step 3.

[0079] Set the training parameters: batch size to 16, use yolov9-c as the initial weights, and training epochs to 200.

[0080] During training, the positive and negative sample matching strategy uses TaskAlign sample matching. The TaskAlignedAssigner matching strategy selects positive samples based on a weighted average of classification and regression scores. t = s a *u b (5)

[0081] For each t, ​​where s is the classification confidence of each anchor point for category t, u is the IoU between each anchor point and the predicted bounding box t, and a and b represent external configuration exponents, multiplying the two yields the alignment metrics. Then, the top k samples are directly selected as positive samples based on the alignment scores.

[0082] The training employed a comprehensive loss function, combining the classification loss (BCE Loss), regression loss (DFL Loss), and CIoU Loss. The specific form of the BCE Loss function is as follows:

[0083] Among them, L BCE (y, y') is the binary crossover loss over the entire dataset; n is the number of samples; y i y is the actual label of the i-th sample, usually 0 or 1 (representing one of the two categories); i' is the probability predicted by the model for the i-th sample, which is usually between 0 and 1.

[0084] The specific form of the DFL Loss function is as follows: L DFL (y i ,y i+1 )=-(i+1-y)log(y i )-(yi)log(y i+1 (7)

[0085] Where y is the actual label, and the variables in the formula satisfy...

[0086] The specific form of the CIoU Loss function is as follows:

[0087] Among them (w) gt h gt (w, h) represent the width and height of the actual bounding box and the predicted bounding box, respectively. Furthermore, CIoU Loss is expressed as: L DIoU =1-CIoU (11)

[0088] The training set and validation set from step 1 are fed into the constructed pavement distress detection model based on YOLOv9 for training according to the set training parameters and processing methods, resulting in a well-converged network model.

[0089] Step 5: Model Evaluation. Based on the improved YOLOv9-based pavement distress detection model obtained after training, the original YOLOv9 network and the improved YOLOv9 network were compared using evaluation metrics such as detection accuracy, return on investment, and average accuracy. The results show that the improved YOLOv9 network performs better in pavement distress detection. The results are shown in Tables 1 and 2, where Table 1 shows the detection results of the original YOLOv9 network and Table 2 shows the detection results of the improved YOLOv9 network.

[0090] Table 1. Detection results of the original YOLOv9 model

[0091] Table 2. Detection results of the improved YOLOv9 model.

[0092] Step 6: As shown in Figure 5, the road surface defect images to be detected are input into the trained road surface defect model based on YOLOv9 for identification, and the output is the road surface defect detection image shown in Figure 6. As shown in the figure, the defect detection image contains the category of each image. The defects are marked with rectangles in the defect detection image and the confidence level is marked.

[0093] In summary, the pavement defect detection method based on improved YOLOv9 provided in the embodiments of the present invention firstly acquires a pavement dataset, cleans the public dataset, and divides the dataset into training, validation, and test sets in an 8:1:1 ratio; secondly, in the head network of the YOLOv9 model, LSKNet modules are added after the RepNCSPELAN4 modules at layers 16, 19, and 22, respectively, to enable the network to better learn and utilize key information in the input data, thereby enhancing the feature expression capability; next, an LSKNet module is added before the SPPELAN layer 11 of the backbone network, enabling the model to better learn and utilize key information in the input data, thereby enhancing the feature expression capability; subsequently, the constructed YOLOv9 model is trained to obtain a model with higher accuracy than the original YOLOv9 network; finally, the trained YOLOv9-based improved model is used to detect the pavement image to be detected to obtain the final detection result.

[0094] Example 2

[0095] This embodiment provides a pavement distress detection device based on an improved YOLOv9 model, including:

[0096] The detection module is used to input the pavement distress image to be detected into the trained pavement distress model based on YOLOv9 for identification and to obtain the detection results.

[0097] The training method for the pavement distress model based on YOLOv9, as described in the detection module, includes:

[0098] Obtain a dataset of road surface damage images and divide it into a training set and a validation set;

[0099] An LSKNet module was added after a specific layer of the YOLOv9 model to construct a pavement distress detection model based on an improved YOLOv9.

[0100] The constructed pavement distress detection model was trained using the training and validation sets to obtain the trained pavement distress model based on YOLOv9.

[0101] Example 3

[0102] This embodiment provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described in Embodiment 1.

[0103] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

[0104] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0105] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more flowchart illustrations and / or one or more block diagrams.

[0106] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams.

[0107] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this disclosure and not to limit its protection scope. Although this disclosure has been described in detail with reference to the above embodiments, those skilled in the art should understand that after reading this disclosure, they can still make various changes, modifications or equivalent substitutions to the specific implementation of the invention, but these changes, modifications or equivalent substitutions are all within the protection scope of the pending claims.

Claims

1. A pavement distress detection method based on an improved YOLOv9 model, characterized in that, include: The images of road surface defects to be detected are input into the trained road surface defect model based on YOLOv9 for identification, and the detection results are obtained. The training method for the pavement distress model based on the improved YOLOv9 includes: Obtain a dataset of road surface damage images and divide it into a training set and a validation set; An LSKNet module was added after a specific layer of the YOLOv9 model to construct a pavement distress detection model based on an improved YOLOv9. The constructed pavement distress detection model was trained using the training and validation sets to obtain the trained pavement distress model based on YOLOv9.

2. The pavement distress detection method based on the improved YOLOv9 model according to claim 1, characterized in that, The method of adding an LSKNet module after a specific layer of the YOLOv9 model to construct an improved pavement distress detection model based on YOLOv9 includes: An LSKNet module is constructed, which includes a large kernel selection module (LK Selection) and a feedforward network module (FFN). The LK Selection module can dynamically adjust the receptive field of the network as needed, and the FFN module is used for channel blending and feature refinement. The LSKNet module is added after the RepNCSPELAN4 module at layers 16, 19, and 22 of the YOLOv9 model's head network, and the LSKNet module is added before the SPPELAN module at layer 11 of the YOLOv9 model's backbone network, resulting in a pavement distress detection model improved based on YOLOv9.

3. The pavement distress detection method based on the improved YOLOv9 model according to claim 2, characterized in that, The LK Selection module includes a fully connected layer sequence, a GELU activation function layer, a core LSK layer, and a second fully connected layer.

4. The pavement distress detection method based on the improved YOLOv9 model according to claim 2, characterized in that, The FFN module includes a sequence of fully connected layers, a depthwise convolution, a GELU activation function layer, and a second fully connected layer.

5. The pavement distress detection method based on the improved YOLOv9 model according to claim 1, characterized in that, The road surface defect image dataset is derived from the RDD2020 competition dataset.

6. The pavement distress detection method based on the improved YOLOv9 model according to claim 1, characterized in that, The pavement distress image dataset contains multiple categories of pavement distress, namely Longitudinal Cracks (D00), Transverse Cracks (D10), Alligator Cracks (D20), and Potholes (D40).

7. The pavement distress detection method based on the improved YOLOv9 model according to claim 1, characterized in that, The test results include information on the type and location of road surface defects.

8. The pavement distress detection method based on the improved YOLOv9 model according to claim 1, characterized in that, The method further includes: dividing a test set from the road surface distress image dataset, and using the test set to evaluate the detection accuracy of the trained road surface distress detection model based on YOLOv9.

9. A pavement distress detection device based on an improved YOLOv9 model, characterized in that, include: The detection module is used to input the pavement distress image to be detected into the trained pavement distress model based on YOLOv9 for identification and to obtain the detection results. The training method for the pavement distress model based on the improved YOLOv9 in the detection module includes: Obtain a dataset of road surface damage images and divide it into a training set and a validation set; An LSKNet module was added after a specific layer of the YOLOv9 model to construct a pavement distress detection model based on an improved YOLOv9. The constructed pavement distress detection model was trained using the training and validation sets to obtain the trained pavement distress model based on YOLOv9.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When executed by a processor, the program implements the steps of the method described in any one of claims 1-8.

Images