Highway rapid detection image pavement disease detection method based on improved YOLOv8 network

Abstract

The application provides a highway rapid detection image pavement disease detection method based on an improved YOLOv8 network, and the method comprises the following steps: constructing a highway rapid detection image pavement disease image labeling data set; constructing five basic models, namely Faster-RCNN-RSDD-HRII, SSD-RSDD-HRII, Retina-Net-RSDD-HRII, YOLOv5-RSDD-HRII and YOLOv8-RSDD-HRII, for highway rapid detection image pavement disease detection; through comparative experiments and cross-validation experiments, it is concluded that the YOLOv8-RSDD-HRII model is more suitable for the pavement disease basic detection network model under the highway rapid detection image scene; considering that the highway rapid detection image pavement disease belongs to a small type target, and the detection rate of the YOLOv8-RSDD-HRII model test result is relatively low, therefore, on the basis of the existing YOLOv8 network model, the improved model YOLOv8-SODL-RSDD-HRII and YOLOv8-CBAM-RSDD-HRII for highway rapid detection image small pavement disease detection are constructed by improving the internal structure of the network; through comparative experiments and cross-validation experiments, it is concluded that the YOLOv8-CBAM-RSDD-HRII model is more suitable for the small pavement disease detection under the highway rapid detection image scene, and the highway rapid detection image pavement disease can be more accurately identified, so that the highway maintenance efficiency is effectively improved.

Description

Technical Field

[0001] This invention belongs to the fields of intelligent transportation and road maintenance, and specifically relates to a method for detecting road surface defects in highway rapid inspection images based on an improved YOLOv8 network. Background Technology

[0002] With the continuous improvement of my country's highway network and the expansion of road scale, more and more road maintenance problems have emerged, the most common of which is road surface defects. After the road surface is damaged, repeated vehicle traffic will cause serious damage to the road and even damage the roadbed. Therefore, the maintenance and management of road surface is particularly important. If road surface defects can be detected and reported in a timely manner, the efficiency of highway maintenance in my country can be effectively improved, maintenance costs can be reduced, and the safety and service life of my country's highways can be improved.

[0003] In the field of highway pavement distress detection, traditional manual or semi-automated methods mainly rely on road maintenance personnel for exploration, measurement, and evaluation analysis. This method is time-consuming, inaccurate, and susceptible to human error, increasing the safety of maintenance workers during on-site operations. However, with the continuous development and innovation of computer science and technology, machine learning and deep learning technologies can automatically learn and detect different types of pavement distress by training on large amounts of highway pavement distress image data. This improves the accuracy and speed of pavement distress detection. Furthermore, video sensors and other equipment can be installed on vehicles to collect pavement data in real time, further enhancing the timeliness of pavement distress detection. Therefore, this invention proposes a rapid image-based pavement distress detection method for highways based on an improved YOLOv8 network. Summary of the Invention

[0004] Purpose of the invention: In order to overcome the shortcomings of the existing technology, this invention provides a method for detecting pavement defects in highway rapid inspection images based on an improved YOLOv8 network. Based on the original YOLOv8 network model, the internal structure is improved to construct an improved network model for detecting minor pavement defects in highway rapid inspection images, providing technical support for the identification and classification of pavement defect information.

[0005] Technical Solution: To achieve the above objectives, this invention provides a method for detecting pavement defects in highway rapid detection images based on an improved YOLOv8 network, comprising the following steps:

[0006] S1: Construct a dataset of labeled road surface defects from rapid highway inspection images;

[0007] S2: Construct five basic models for road surface defect detection in highway fast detection images: Faster-RCNN-RSDD-HRII, SSD-RSDD-HRII, Retina-Net-RSDD-HRII, YOLOv5-RSDD-HRII, and YOLOv8-RSDD-HRII.

[0008] S3: Based on the constructed highway rapid detection pavement disease image annotation dataset, conduct comparative experiments to find a pavement disease basic detection network model that is more suitable for highway rapid detection image scenarios;

[0009] S4: Construct an improved model YOLOv8-SODL-RSDD-HRII for detecting minor road surface defects in rapid highway images;

[0010] S5: Construct an improved model YOLOv8-CBAM-RSDD-HRII for detecting minor road surface defects in rapid highway images;

[0011] S6: Based on the constructed highway rapid detection pavement defect image annotation dataset, conduct comparative experiments to find a network model that is more suitable for detecting small pavement defects in highway rapid detection image scenarios.

[0012] Furthermore, the specific method for constructing the highway rapid detection image pavement defect image annotation dataset in step S1 is as follows:

[0013] S1-1: First, acquire the original images of the road surface for rapid inspection. The collected image data are images taken by the data collection vehicle on the Baolong G5XY section of the highway from August 18 to 21, 2022, at a constant speed, with a fixed shooting time interval, and the angle between the area array video sensor and the horizontal ground is 30°. The resolution of the collected road surface images for rapid inspection is 2064×1544.

[0014] S1-2: From the 18,713 original highway pavement images collected, images with obvious pavement distress characteristics were selected, mainly including transverse cracks and longitudinal cracks. Images with unclear pavement distress characteristics or overly complex and messy pavement distress characteristics were deleted, resulting in a total of 4,000 effective highway pavement distress images.

[0015] S1-3: LabelImg software was used to annotate the road surface defects images of the highway rapid detection. A segmented annotation method was adopted to ensure that the road surface defect information in each segment was consistent. The dataset label name is: {Transverse,Longitudinal}.

[0016] S1-4: After annotation, a highway pavement defect annotation dataset of 4000 images based on highway rapid detection images was obtained, including 11892 transverse cracks and 3495 longitudinal cracks, with a total of 15387 labels. The 4000 annotated highway pavement images were then randomly divided into training, validation, and test sets in a ratio of 8:1:1.

[0017] Furthermore, the specific steps for constructing the five basic models for road surface distress detection in highway rapid detection images in step S2 are as follows:

[0018] S2-1: Construct a Faster-RCNN-RSDD-HRII model for detecting pavement defects in highway rapid detection images;

[0019] S2-2: Constructing the SSD-RSDD-HRII model for pavement distress detection in rapid highway image analysis;

[0020] S2-3: Construct a Retina-Net-RSDD-HRII model for road surface distress detection in rapid highway image analysis;

[0021] S2-4: Construct a YOLOv5-RSDD-HRII model for detecting pavement defects in highway rapid detection images;

[0022] S2-5: Construct a YOLOv8-RSDD-HRII model for detecting pavement defects in highway rapid detection images.

[0023] Furthermore, in step S3, a comparative experimental study of the basic model is conducted based on the constructed highway rapid detection pavement defect image annotation dataset. The main contents are as follows:

[0024] S3-1: Select experimental evaluation indicators. The harmonic mean of precision and recall, F1-Score, is used as the evaluation indicator for road surface defect detection in highway rapid detection images, and the average time required to detect one image (in milliseconds) is used as the real-time indicator.

[0025] S3-2: During network model training, observe the loss curve of the network model. If the loss curve drops to its minimum and remains stable, it indicates that the network model has converged. After multiple debugging and training sessions, observe the changes in the loss function curves of each model to determine if the loss function curves of each model have converged. This conclusion can be applied to the test set for statistical experimental evaluation metrics.

[0026] S3-3: Based on the constructed highway rapid detection pavement defect image dataset, a comparative experiment was conducted. The experimental results show that the overall detection performance of the YOLOv8-RSDD-HRII model is superior to the other four models. Its Precision, Recall, F1-score, and real-time performance indicators are 98.81%, 88.89%, 93.59%, and 33.3ms, respectively. This indicates that the YOLOv8-RSDD-HRII model is the most suitable basic detection model network for pavement defects in highway rapid detection image scenarios. However, the detection rate of the test results is low, only 88.89%. The possible reasons for this result and the improvement measures are as follows:

[0027] ①Possible causes: The proportion of road surface defects in the rapid inspection images of highways is very small. Transverse and longitudinal cracks are very small line segments in the original image. In addition, the area array video sensor of the acquisition vehicle is at a certain angle to the ground, so the rapid inspection images of highways often have the characteristic of near objects being larger and far objects being smaller, and the imaging has strong distortion.

[0028] ② Improvement measures: Considering that road surface defects in highway rapid detection images are small-scale targets, a network model for detecting small road surface defects in highway rapid detection images was constructed by improving the internal structure of the existing YOLOv8-RSDD-HRII network model, and comparative experimental studies were carried out.

[0029] Furthermore, in step S4, an improved model YOLOv8-SODL-RSDD-HRII (YOLOv8 with Small Object Detection Layer for RoadSurface Disease Detection using Highway Rapid Inspection Images, YOLOv8-SODL-RSDD-HRII) is constructed. Specifically, based on the original YOLOv8-RSDD-HRII network model, a new detection layer with a scale of 160×160×21 is added to the Head layer structure. Correspondingly, the Backbone layer and Neck layer of the network also add corresponding Conv convolutional layers and C2f convolutional modules to cooperate with this new small object detection layer. The improved model YOLOv8-SODL-RSDD-HRII outputs four feature maps at scales of 160×160×21, 80×80×21, 40×40×21, and 20×20×21. The Head part outputs feature maps at four × two = eight scales for classification and regression. The front-end feature extraction part and subsequent processes are the same as the YOLOv8 network workflow. The class prediction branches and bounding box prediction branches at these four different scales are concatenated and subjected to a series of processes such as dimensional transformation, and finally the pavement disease detection information of the highway fast detection image of the improved model YOLOv8-SODL-RSDD-HRII is obtained.

[0030] Furthermore, in step S5, an improved model YOLOv8-CBAM-RSDD-HRII (YOLOv8 with Convolutional Block Attention Module for Road Surface Disease Detection using Highway Rapid Inspection) is constructed for detecting minor road surface defects in highway rapid inspection images. Images (YOLOv8-CBAM-RSDD-HRII) are implemented by adding a CBAM module after the C2f module in the backbone of the original YOLOv8-RSDD-HRII network model. The parameters of the CBAM module after the first C2f module are 128×7, the second is 256×7, and the third is 512×7, depending on the feature map. The first variable represents the number of channels in the feature map of that layer, and the second variable represents the kernel size of the spatial attention mechanism in the CBAM attention module. For example, a CBAM module with parameters of 128×7 has a kernel size of 7×7 and 128 channels. Other parts are consistent with the YOLOv8-RSDD-HRII model.

[0031] Furthermore, in step S6, a comparative experimental analysis of the improved model is conducted based on the constructed highway rapid detection pavement defect image annotation dataset. The main contents are as follows:

[0032] S6-1: The harmonic mean F1-Score of precision and recall is still used as the evaluation index for detecting minor road surface defects in highway rapid detection images, and the average time required to detect one image (in milliseconds) is used as the real-time index.

[0033] S6-2: During network model training, observe the loss function curve of the network model. If the loss function curve drops to its minimum and remains stable, it indicates that the network model has converged. After multiple debugging and training sessions, observe the changes in the loss function curves of each model to determine if the loss function curves of each model have converged. This conclusion can be applied to the test set for statistical evaluation metrics.

[0034] S6-3: Based on the constructed highway rapid detection pavement defect image dataset, a comparative experiment was conducted. The experimental results show that the overall detection performance of the YOLOv8-CBAM-RSDD-HRII model is better than the other two models. Its Precision, Recall, F1-score and real-time performance indicators are 98.10%, 91.81%, 94.85% and 36.9ms, respectively, indicating that the YOLOv8-CBAM-RSDD-HRII model is more suitable for detecting small pavement defects in highway rapid detection image scenarios. Attached Figure Description

[0035] Figure 1 Schematic diagram of a highway rapid inspection image acquisition vehicle;

[0036] Figure 2 To improve the structure diagram of the YOLOv8-SODL-RSDD-HRII model;

[0037] Figure 3 To improve the structure diagram of the YOLOv8-CBAM-RSDD-HRII model;

[0038] Figure 4 This is a schematic diagram of the CBAM attention module.

[0039] Figure 5 Box plots of the F1-Score indicators for the five basic models;

[0040] Figure 6 Box plots of the F1-Score index for three improved models. Detailed Implementation

[0041] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0042] This invention provides a method for detecting pavement defects in highway rapid detection images based on an improved YOLOv8 network, comprising the following steps:

[0043] S1: Construct a dataset of labeled road surface defects from rapid highway detection images. The specific method is as follows:

[0044] S1-1: First, acquire the original images of the highway surface for rapid inspection. The collected image data consists of images taken by the data acquisition vehicle on the Baolong G5XY section highway from August 18th to 21st, 2022, at a constant speed, with fixed shooting intervals, and the angle between the area array video sensor and the horizontal ground at 30°. The highway rapid inspection image data acquisition vehicle is as follows: Figure 1 As shown, it mainly includes an area array video sensor and storage device, and the resolution of the collected highway rapid inspection road surface images is 2064×1544.

[0045] S1-2: From the 18,713 original highway pavement images collected, images with obvious pavement distress characteristics were selected, mainly including transverse cracks and longitudinal cracks. Images with unclear pavement distress characteristics or overly complex and messy pavement distress characteristics were deleted, resulting in a total of 4,000 effective highway pavement distress images.

[0046] S1-3: LabelImg software was used to annotate the road surface defects images of the highway rapid detection. A segmented annotation method was adopted to ensure that the road surface defect information in each segment was consistent. The dataset label name is: {Transverse,Longitudinal}.

[0047] S1-4: After annotation, a dataset of 4000 highway pavement defect annotations based on highway fast detection images was obtained, including 11892 transverse cracks and 3495 longitudinal cracks, with a total of 15387 labels. The 4000 annotated highway pavement images were then randomly divided into training, validation, and test sets in a ratio of 8:1:1.

[0048] S2: Construct five basic models for road surface distress detection using rapid highway image analysis. The specific steps are as follows:

[0049] S2-1: Construct a Faster-RCNN-RSDD-HRII model for detecting pavement defects in highway rapid detection images;

[0050] S2-2: Constructing the SSD-RSDD-HRII model for pavement distress detection in rapid highway image analysis;

[0051] S2-3: Construct a Retina-Net-RSDD-HRII model for road surface distress detection in rapid highway image analysis;

[0052] S2-4: Construct a YOLOv5-RSDD-HRII model for detecting pavement defects in highway rapid detection images;

[0053] S2-5: Construct a YOLOv8-RSDD-HRII model for detecting pavement defects in highway rapid detection images.

[0054] S3: Based on the constructed highway rapid detection pavement defect image annotation dataset, a comparative experimental study of the basic model was conducted. The main contents are as follows:

[0055] S3-1: Select experimental evaluation indicators. The harmonic mean of precision and recall, F1-Score, is used as the evaluation indicator for road surface defect detection in highway rapid detection images, and the average time required to detect one image (in milliseconds) is used as the real-time indicator.

[0056] S3-2: During network model training, observe the loss curve of the network model. If the loss curve drops to its minimum and remains stable, it indicates that the network model has converged. After multiple debugging and training sessions, observe the changes in the loss function curves of each model to determine if the loss function curves of each model have converged. This conclusion can be applied to the test set for statistical experimental evaluation metrics.

[0057] S3-3: Based on the constructed highway rapid detection pavement defect image dataset, a comparative experiment was conducted. The experimental results show that the overall detection performance of the YOLOv8-RSDD-HRII model is superior to the other four models. Its Precision, Recall, F1-score, and real-time performance indicators are 98.81%, 88.89%, 93.59%, and 33.3ms, respectively. This indicates that the YOLOv8-RSDD-HRII model is the most suitable basic detection model network for pavement defects in highway rapid detection image scenarios. However, the detection rate of the test results is low, only 88.89%. The possible reasons for this result and the improvement measures are as follows:

[0058] ①Possible causes: The proportion of road surface defects in the rapid inspection images of highways is very small. Transverse and longitudinal cracks are very small line segments in the original image. In addition, the area array video sensor of the acquisition vehicle is at a certain angle to the ground, so the rapid inspection images of highways often have the characteristic of near objects being larger and far objects being smaller, and the imaging has strong distortion.

[0059] ② Improvement measures: Considering that road surface defects in highway rapid detection images are small-scale targets, based on the existing YOLOv8-RSDD-HRII network model, a network model for detecting small road surface defects in highway rapid detection images is constructed by improving the internal structure of the network, and comparative experimental studies are carried out.

[0060] S4: Construct an improved model, YOLOv8-SODL-RSDD-HRII (YOLOv8 with Small Object Detection Layer for Road Surface Disease Detection using Highway Rapid Inspection Images), for detecting small road surface defects in highway rapid inspection images. Its structure is as follows: Figure 2As shown, the specific approach is as follows: Based on the original YOLOv8-RSDD-HRII network model, a new detection layer with a scale of 160×160×21 is added to the Head layer structure. Correspondingly, Conv convolutional layers and C2f convolutional modules are added to the Backbone and Neck layers of the network to cooperate with this new small target detection layer. The improved model YOLOv8-SODL-RSDD-HRII outputs four feature maps with scales of 160×160×21, 80×80×21, 40×40×21, and 20×20×21. The Head part outputs feature maps of 4×2=8 scales for classification and regression. The front-end feature extraction part and subsequent processes are the same as the YOLOv8 network workflow. The class prediction branches and bounding box prediction branches of these four different scales are concatenated and subjected to a series of processes such as dimensionality transformation, and finally the road surface defect detection information of the highway fast detection image of the improved model YOLOv8-SODL-RSDD-HRII is obtained.

[0061] S5: Construct an improved model YOLOv8-CBAM-RSDD-HRII (YOLOv8 with Convolutional Block Attention Module for Road SurfaceDisease Detection using Highway Rapid Inspection Images) for detecting minor road surface defects in highway rapid inspection images. Its structure is as follows: Figure 3 As shown, the specific implementation is as follows: The CBAM module consists of two sub-modules: the Channel Attention Module (CAM) and the Spatial Attention Module (SAM). The CBAM module generates new sub-feature maps along both the channel and spatial dimensions. These sub-feature maps are then multiplied by the original feature map to obtain the optimized new feature map, as shown below. Figure 4As shown, based on the original YOLOv8-RSDD-HRII network model, a CBAM module is added after the C2f module in its backbone. The parameters of the CBAM module after the first C2f module are 128×7, the second C2f module is 256×7, and the third C2f module is 512×7, depending on the feature map. The first variable is the number of channels in the feature map of that layer, and the second variable is the size of the convolution kernel of the spatial attention mechanism in the CBAM attention module. For example, the CBAM module with parameters of 128×7 has a convolution kernel size of 7×7 and 128 channels. The other parts are consistent with the YOLOv8-RSDD-HRII model.

[0062] S6: Based on the constructed highway rapid detection pavement defect image annotation dataset, comparative experimental analysis of the improved model is conducted, the main contents of which are as follows:

[0063] S6-1: The harmonic mean F1-Score of precision and recall is still used as the evaluation index for detecting minor road surface defects in highway rapid detection images, and the average time required to detect one image (in milliseconds) is used as the real-time index.

[0064] S6-2: During network model training, observe the loss function curve of the network model. If the loss function curve drops to its minimum and remains stable, it indicates that the network model has converged. After multiple debugging and training sessions, observe the changes in the loss function curves of each model to determine if the loss function curves of each model have converged. This conclusion can be applied to the test set for statistical evaluation metrics.

[0065] S6-3: Based on the constructed highway rapid detection pavement defect image dataset, a comparative experiment was conducted. The experiment showed that the overall detection performance of the YOLOv8-CBAM-RSDD-HRII model is better than the other two models. Its Precision, Recall, F1-score and real-time performance indicators are 98.10%, 91.81%, 94.85% and 36.9ms, respectively, indicating that the YOLOv8-CBAM-RSDD-HRII model is more suitable for detecting small pavement defects in highway rapid detection image scenarios.

[0066] To verify the superiority of the YOLOv8-RSDD-HRII model, a comparative experiment was conducted in this case study. Based on a pre-constructed highway pavement defect annotation dataset of rapid highway image detection, the dataset was randomly divided into training, validation, and test sets in an 8:1:1 ratio. Five basic models—Faster-RCNN-RSDD-HRII, SSD-RSDD-HRII, Retina-Net-RSDD-HRII, YOLOv5-RSDD-HRII, and YOLOv8-RSDD-HRII—were used to perform perceptual recognition experiments on the standard dataset. The neural network was then trained, and the detection results for the test set were obtained, as shown in Tables 10-12. It can be seen that the overall detection performance of the YOLOv8-RSDD-HRII model is superior to the other four models, indicating that the YOLOv8-RSDD-HRII model is more suitable for basic pavement defect detection in rapid highway image detection scenarios.

[0067] Table 10 Experimental Results of Five Basic Models

[0068]

[0069]

[0070] Table 11 Results of transverse crack detection for five basic models

[0071]

[0072] Table 12. Longitudinal crack detection results for five basic models.

[0073]

[0074] To further demonstrate the effectiveness of the YOLOv8-RSDD-HRII model, 20 cross-validation experiments were conducted. In each experiment, 100 images of road surface defects were randomly selected from the road rapid detection image test sample library for testing. The detection accuracy of the five models was statistically analyzed, and their F1-Score box plots are shown below. Figure 5 As shown in the figure, the YOLOv8-RSDD-HRII model has the highest detection accuracy and a smaller variation in the distribution range of its F1-Score value. The detection results of YOLOv5-RSDD-HRII and Faster-RCNN-RSDD-HRII are similar. The Retina-Net-RSDD-HRII model has a larger distribution range of its F1-score value, indicating that the model is less stable.

[0075] To verify the superiority of the improved model, comparative experiments were also conducted. Based on the constructed highway rapid detection image pavement defect annotation dataset, and randomly divided into training, validation, and test sets in an 8:1:1 ratio, three highway rapid detection image pavement defect detection models—YOLOv8-RSDD-HRII, YOLOv8-SODL-RSDD-HRII, and YOLOv8-CBAM-RSDD-HRII—were trained respectively. The detection results of different models on the test set were compared and analyzed, and the results are shown in Tables 13-15. It can be seen that the overall detection performance of the YOLOv8-CBAM-RSDD-HRII model is better than the other two models, indicating that the YOLOv8-CBAM-RSDD-HRII model is more suitable for detecting small pavement defects in highway rapid detection image scenarios.

[0076] Table 13 Experimental results of the three models

[0077]

[0078]

[0079] Table 14 Results of transverse crack detection for three models

[0080]

[0081] Table 15. Results of longitudinal crack detection using three models

[0082]

[0083] To further demonstrate the effectiveness of the YOLOv8-CBAM-RSDD-HRII model, 20 cross-validation experiments were conducted. In each experiment, 100 images of road surface defects detected by rapid road inspection were randomly selected from the test sample library for testing. The detection accuracy of the three models was statistically analyzed, and their F1-Score box plots are shown below. Figure 6 As shown in the figure, the YOLOv8-CBAM-RSDD-HRII model has the highest detection accuracy and a smaller F1-Score distribution range.

[0084] The technical means disclosed in this invention are not limited to those disclosed in the above embodiments, but also include technical solutions composed of any combination of the above technical features. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications are also considered within the scope of protection of this invention.

Claims

1. A method for detecting pavement defects in highway images based on an improved YOLOv8 network, characterized in that: Includes the following steps: S1: Construct a dataset of labeled road surface defects from rapid highway inspection images; S2: Construct five basic models for road surface defect detection in highway fast detection images: Faster-RCNN-RSDD-HRII, SSD-RSDD-HRII, Retina-Net-RSDD-HRII, YOLOv5-RSDD-HRII, and YOLOv8-RSDD-HRII. S3: Based on the constructed highway rapid detection pavement defect image annotation dataset, conduct comparative experiments on basic models to find a pavement defect basic detection network model that is more suitable for highway rapid detection image scenarios. S4: Construct an improved model, YOLOv8-SODL-RSDD-HRII, for detecting minor road surface defects in highway rapid image detection. Specifically, based on the existing YOLOv8-RSDD-HRII network model, a new detection layer with a scale of 160×160×21 is added to the Head layer. Correspondingly, Conv and C2f convolutional modules are added to the Backbone and Neck layers to cooperate with this new small target detection layer. The improved model YOLOv8-SODL- The RSDD-HRII output feature maps have four scales: 160×160×21, 80×80×21, 40×40×21, and 20×20×21. The Head part outputs feature maps of 4×2=8 scales for classification and regression. The front-end feature extraction part and subsequent processes are the same as the YOLOv8 network. The class prediction branch and bounding box prediction branch of these four different scales are concatenated and a series of dimensional transformations are performed to finally obtain the improved model YOLOv8-SODL-RSDD-HRII. S5: Construct an improved model YOLOv8-CBAM-RSDD-HRII for detecting minor road surface defects in highway rapid detection images. Specifically, based on the original YOLOv8-RSDD-HRII network model, add a CBAM module after the C2f module in its backbone. Depending on the feature map, the CBAM module after the first C2f module has parameters of 128×7, the second C2f module has parameters of 256×7, and the third C2f module has parameters of 512×7. The first variable is the number of channels in the feature map of that layer, and the second variable is the size of the convolution kernel of the spatial attention mechanism in the CBAM attention module. The CBAM module with parameters of 128×7 has a convolution kernel size of 7×7 and 128 channels. The other parts are consistent with the YOLOv8-RSDD-HRII model. S6: Based on the constructed highway rapid detection pavement defect image annotation dataset, conduct comparative experiments on the improved model to find a network model that is more suitable for detecting small pavement defects in highway rapid detection image scenarios.

2. The method for detecting pavement defects in highway rapid detection images based on an improved YOLOv8 network according to claim 1, characterized in that: The specific method for constructing the highway rapid detection image pavement defect image annotation dataset in step S1 is as follows: S1-1: First, acquire the original image of the road surface for rapid inspection. The acquired image data is the image taken by the data acquisition vehicle when it is traveling at a constant speed on a certain section of the road, the shooting time interval is fixed, and the angle between the area array video sensor and the horizontal ground is 30°. The resolution of the acquired road surface image for rapid inspection is 2064×1544. S1-2: From the 18,713 original highway pavement images collected, images with obvious pavement distress characteristics were selected, including transverse and longitudinal cracks. Images with unclear pavement distress characteristics or overly complex and messy pavement distress characteristics were deleted. Finally, a total of 4,000 effective highway pavement distress images were obtained. S1-3: LabelImg software was used to annotate the road surface defects images of the highway rapid detection. A segmented annotation method was adopted to ensure that the road surface defect information in each segment was consistent. The dataset label name is: {Transverse,Longitudinal}. S1-4: After the annotation was completed, a total of 4,000 images of highway pavement disease annotation dataset based on highway fast detection images were obtained, including 11,892 transverse cracks, 3,495 longitudinal cracks, and a total of 15,387 labels. The 4,000 labeled highway pavement images were then randomly divided into training, validation, and test sets in a ratio of 8:1:

1.

3. The method for detecting pavement defects in highway rapid detection images based on an improved YOLOv8 network according to claim 1, characterized in that: The specific steps for constructing the five basic models for road surface defect detection in highway rapid detection images in step S2 are as follows: S2-1: Construct the Faster-RCNN model Faster-RCNN-RSDD-HRII for road surface defect detection in highway rapid detection images. Its model structure parameters are shown in Table 1. In Table 1, Resolution is the size of the variable in each Stage, Channels is the number of channels of the variable in each Stage, Layers is the number of times the Operator is repeated in each Stage, Conv is the convolution operation, which includes the ReLU activation process, Kernels is the size of the convolution kernel for the convolution operation or pooling operation, Max Pool is the maximum pooling operation on the variable, and FC is the fully connected layer. The first fully connected layer outputs the category information, and the second fully connected layer outputs the vertex coordinate information of the detection box for each category. Table 1. Structure parameters of the Faster-RCNN-RSDD-HRII model S2-2: Construct the SSD model SSD-RSDD-HRII for road surface defect detection in highway rapid detection images. The model structure parameters are shown in Table 2. In Table 2, Resolution is the size of the variable in each Stage, Channels is the number of channels of the variable in each Stage, Layers is the number of times the Operator is repeated in each Stage, Conv is the convolution operation, Kernels is the size of the convolution kernel for the convolution or pooling operation, Max Pool is the maximum pooling operation on the variable, and Detect_X is the convolution kernel of the predictor with the number of channels being 24, 36, 36, 36, 24, and 24, which are related to the detection category. Table 2 SSD-RSDD-HRII Model Structure Parameters S2-3: Construct the Retina-Net-RSDD-HRII model for road surface defect detection in highway rapid detection images. Its model structure parameters are shown in Table 3. In the table, Resolution is the size of the variable in each Stage, Channels is the number of channels of the variable in each Stage, Kernels is the kernel size of each Stage, Layers is the number of times the Operator is repeated in each Stage, Conv includes Convolution, Batch Normalization, and ReLU activation. For example, Conv_1 in the table is the convolution operation on the input variable using a 3×3 kernel, and Maxpool is the max pooling operation on the input variable; Res is the residual structure; Feature is the feature layer output by the network, which includes 5 layers: Clsss_subnet is the class predictor, and Box_subnet is the bounding box regression parameter predictor. Table 3. Retina-Net-RSDD-HRII Model Structure Parameters S2-4: Construct the YOLOv5 model YOLOv5-RSDD-HRII for road surface defect detection in highway rapid detection images. The model structure parameters are shown in Table 4. In Table 4, Resolution is the scale of the variables in each Stage, Channels is the number of channels for the variables in each Stage, Layers is the number of times the Operator operation is repeated in each Stage, Kernels is the size of the convolution kernel for the convolution operation, CBL is to perform convolution operation on the variables of the layer first, then perform standardization operation, and use the LeakyReLU activation function for activation; C3 is the CSP1_X module, SPP 5×5 is the convolution kernel size of 5×5 in the SPP structure, Unsample is the upsampling operation on the variables, Concat is the concatenation operation on the variables, and the concatenation level is the C3 module connected to it and the C3 module of the same size in the backbones. Detect is the output module, and the output dimension is related to the number of target categories. In this paper, the road surface defects in the highway rapid detection images are divided into two categories, so the number of output channels is 3×(2+5)=21. Table 4. Structural parameters of the YOLOv5-RSDD-HRII model S2-5: Construct the YOLOv8 network YOLOv8-RSDD-HRII for pavement defect detection in highway rapid image inspection. Its model structure parameters are shown in Table 5. Here, Resolution is the scale of the variables in each Stage, Channels is the number of channels for each Stage variable, Layers is the number of times the Operator operation is repeated in each Stage, Kernels is the size of the convolution kernel for the convolution operation, CBL is the layer where convolution is performed first, followed by normalization and activation using the LeakyReLU activation function; C2f is the C2f network structure layer, SPPF... 5×5 refers to the 5×5 kernel size in the SPPF structure. Unsample is an upsampling operation on the variables. Concat is a concatenation operation on the variables. The concatenation level is the C2f module connected to it and the C2f module of the same size in the Backbone. Detect_X is the output module. The output dimension is related to the number of target categories. In the road surface image of the highway quick detection in this paper, the road surface defects are divided into two categories. Therefore, the number of output channels is 3×(2+5)=21. Table 5. Structural parameters of the YOLOv8-RSDD-HRII model 4. The method for detecting pavement defects in highway rapid detection images based on an improved YOLOv8 network according to claim 1, characterized in that: In step S3, a comparative experimental study of the basic model is conducted based on the constructed highway rapid detection pavement defect image annotation dataset. The main contents are as follows: S3-1: Select experimental evaluation indicators. The harmonic mean of accuracy and recall, F1-Score, is used as the evaluation indicator for pavement defects detection in highway rapid detection images, and the average time required to detect one image is used as the real-time indicator. S3-2: When training the network model, observe the loss curve of the network model. If the loss curve drops to the minimum value and remains stable, it indicates that the network model has converged. After multiple debugging and training, the basic parameters for training five models, Faster-RCNN-RSDD-HRII, SSD-RSDD-HRII, Retina-Net-RSDD-HRII, YOLOv5-RSDD-HRII, and YOLOv8-RSDD-HRII, were finally determined, as shown in Table 6. The changes in the loss function curve of each model were observed, and it was found that the loss function curve of each model converged and could be applied to the statistical experimental evaluation index on the test set. Table 6 Training parameters for five models S3-3: Based on the constructed highway rapid detection pavement defect image dataset, a comparative experiment was conducted. The experimental results show that the overall detection performance of the YOLOv8-RSDD-HRII model is better than the other four models. Its Precision, Recall, F1-score and real-time performance indicators are 98.81%, 88.89%, 93.59% and 33.3ms, respectively. This indicates that the YOLOv8-RSDD-HRII model is more suitable for the basic detection network model of pavement defects in the scenario of highway rapid detection images. However, the detection rate of the test results is low, only 88.89%.

5. The method for detecting pavement defects in highway rapid detection images based on an improved YOLOv8 network according to claim 4, characterized in that: The possible reasons and improvement measures for the low detection rate of the YOLOv8-RSDD-HRII model for road surface defects in step S3-3 are as follows: ①Possible causes: The proportion of road surface defects in the rapid inspection images of highways is very small. Transverse and longitudinal cracks are very small line segments in the original image. In addition, the area array video sensor of the acquisition vehicle is at a certain angle to the ground, so the rapid inspection images of highways often have the characteristic of near objects being larger and far objects being smaller, and the imaging has strong distortion. ② Improvement measures: Considering that road surface defects in highway rapid detection images are small-scale targets, a network model for detecting small road surface defects in highway rapid detection images was constructed by improving the internal structure of the existing YOLOv8-RSDD-HRII network model, and comparative experimental studies were carried out.

6. The method for detecting pavement defects in highway rapid detection images based on an improved YOLOv8 network according to claim 1, characterized in that: In step S4, an improved model YOLOv8-SODL-RSDD-HRII for detecting minor road surface defects in highway rapid detection images is constructed. The final model structure parameters are shown in Table 7. Here, Resolution is the scale of the variable in each Stage, Channels is the number of channels for the variable in each Stage, Layers is the number of times the Operator operation is repeated in each Stage, Kernels is the size of the convolution kernel for the convolution operation, CBL is the layer where convolution is performed first, followed by normalization and activation using the LeakyReLU activation function; C2f is the C2f network structure layer, and SPPF... 5×5 refers to the 5×5 kernel size in the SPPF structure. Unsample is an upsampling operation on the variables. Concat is a concatenation operation on the variables. The concatenation level is the C2f module connected to it and the C2f module of the same size in the Backbone. Detect_X is the output module. The output dimension is related to the number of target categories. In the road surface image of the highway quick detection in this paper, the road surface defects are divided into two categories. Therefore, the number of output channels is 3×(2+5)=21. Table 7. Structural parameters of the YOLOv8-SODL-RSDD-HRII model.

7. The method for detecting pavement defects in highway rapid detection images based on an improved YOLOv8 network according to claim 1, characterized in that: In step S5, an improved model YOLOv8-CBAM-RSDD-HRII for detecting minor road surface defects in highway rapid detection images is constructed. The final model structure parameters are shown in Table 8. Here, Resolution is the scale of the variables in each Stage, Channels is the number of channels for each Stage variable, Layers is the number of times the Operator operation is repeated in each Stage, Kernels is the size of the convolution kernel for the convolution operation, CBL indicates that the variables in this layer are first convolved, then normalized, and activated using the LeakyReLU activation function; C2f is the C2f network structure layer, Kernel7×7 of the CBAM module indicates that the convolution kernel size in the CBAM module is 7×7, and SPPF... 5×5 refers to the 5×5 kernel size in the SPPF structure. Unsample is an upsampling operation on the variables. Concat is a concatenation operation on the variables. The concatenation level is the C2f module connected to it and the C2f module of the same size in the Backbone. Detect_X is the output module. The output dimension is related to the number of target categories. In the road surface image of the highway quick detection in this paper, the road surface defects are divided into two categories. Therefore, the number of output channels is 3×(2+5)=21. Table 8. Structural parameters of the YOLOv8-CBAM-RSDD-HRII model 8. The method for detecting pavement defects in highway rapid detection images based on an improved YOLOv8 network according to claim 1, characterized in that: In step S6, a comparative experimental analysis of the improved model is conducted based on the constructed highway rapid detection pavement defect image annotation dataset. The main contents are as follows: S6-1: The harmonic mean F1-Score of Precision and Recall is still used as the evaluation index for detecting minor road surface defects in highway rapid detection images, and the average time required to detect one image is used as the real-time index. S6-2: During network model training, observe the loss function curve of the network model. If the loss function curve drops to the minimum value and remains stable, it indicates that the network model has converged. After multiple debugging and training, the basic parameters for training the three improved models YOLOv8-RSDD-HRII, YOLOv8-SODL-RSDD-HRII, and YOLOv8-CBAM-RSDD-HRII are finally determined as shown in Table 9. Observe the changes in the loss function curves of each model to conclude that the loss function curves of each model have converged and can be applied to the statistical experimental evaluation index on the test set. Table 9 Training parameters for three improved models S6-3: Based on the constructed highway rapid detection pavement defect image dataset, a comparative experiment was conducted. The experimental results show that the overall detection performance of the YOLOv8-CBAM-RSDD-HRII model is better than the other two models. Its Precision, Recall, F1-score and real-time performance indicators are 98.10%, 91.81%, 94.85% and 36.9ms, respectively, indicating that the YOLOv8-CBAM-RSDD-HRII model is more suitable for detecting small pavement defects in highway rapid detection image scenarios.

Images