Bridge unmanned aerial vehicle disease identification and evaluation method based on YOLOv8 model

Abstract

The invention provides a bridge unmanned aerial vehicle disease identification and evaluation method based on a YOLOv8 model, and the method specifically comprises the following steps: S1, collecting disease image information, marking a data set as a yolk training format through Label, and obtaining a marked data set; s2, building a YOLOv8 depth model basic framework, modifying the model for small sample disease types, adding a larger-size detection head, and introducing a CBAM attention mechanism into the YOLOv8 model; s3, key indexes of mAPs and detection rates of five models of YOLOv8n, YOLOv8s, YOLOv8m, YOLOv81 and YOLOv8x are tested respectively, and a qualitative evaluation algorithm of the bridge structure is established; and S4, carrying out inspection on the bridge structure, carrying out real-time detection on the inspection video data stream by using the detection model, and carrying out qualitative evaluation through a qualitative evaluation algorithm of the bridge structure.

Description

Technical Field

[0001] The present invention relates to the field of intelligent operation and maintenance of infrastructure, and in particular to a bridge drone disease identification and assessment method based on a YOLOv8 model. Background Art

[0002] The safety of bridge structures is directly related to the safety of people's lives and property, so the detection and identification of bridge defects is of great significance. Traditional bridge inspection methods rely on artificial vision, which has problems such as low efficiency, high cost, and poor safety. With the development of drone technology, its low cost and high flexibility provide new solutions for bridge inspection. Drones can remotely, non-contactly, and automatically inspect bridge structures, especially showing unique advantages in the inspection of high-rise or large bridges.

[0003] The progress of deep learning technology, especially the application of convolutional neural network (CNN) in the field of image recognition, has greatly improved the accuracy of image recognition systems carried by drones. For example, the bridge disease detection system based on the YOLOv5 algorithm and the intelligent crack recognition method based on the Mask R-CNN algorithm collect disease images through drone cruising and use image cutting and splicing technology to realize automatic detection and positioning of diseases. This makes the drone method have important scientific and engineering application value in bridge disease detection. However, for some smaller diseases, drones still have omissions in real-time capture and detection, and the detection efficiency and recognition accuracy need to be improved.

[0004] Therefore, the technical problem that needs to be solved urgently is how to further improve the detection efficiency and the accuracy of disease identification, so as to facilitate the early detection of bridge diseases and provide timely maintenance for them, and meet the needs of bridge safety maintenance. Summary of the invention

[0005] The present invention is conducted to solve the above-mentioned problems, and its purpose is to provide a bridge drone disease identification and assessment method based on the YOLOv8 model.

[0006] The present invention provides a bridge drone disease identification and assessment method based on the YOLOv8 model, which has the following characteristics and specifically includes the following steps: S1, using a drone high-definition camera to take photos of key components of the highway bridge structure, including the bridge deck pavement and the superstructure, collect disease image information, and use Labelme to annotate the data set into the YOLO training format to obtain an annotated data set; S2, using VS Code builds the basic framework of the YOLOv8 deep model. When the type of disease is a small sample disease type, the model is modified by adding a larger detection head and introducing the CBAM attention mechanism into the YOLOv8 model, thereby improving the detection accuracy and forming a deep model framework suitable for bridge drone disease recognition; S3, based on the labeled data set and the built deep model framework suitable for bridge drone disease recognition, the key indicators of mAP and detection rate of five models, YOLOv8n, YOLOv8s, YOLOv8m, YOLOv8l and YOLOv8x, are tested respectively, and the appropriate detection model and number of training epochs are selected. The detection results of the drone detection video stream are classified based on the trained model, and a qualitative evaluation algorithm for the bridge structure is established; and S4, the bridge structure is inspected by a drone, the inspection video data stream is detected in real time using the detection model, and a qualitative evaluation is performed through the qualitative evaluation algorithm of the bridge structure.

[0007] In the bridge drone disease identification and assessment method based on the YOLOv8 model provided by the present invention, it can also have the following characteristics: wherein the diseases in step S1 include the following types: longitudinal seams, subsidence, bulges, rutting, potholes, looseness, spalling, oil overflow, repairs and structural cracks.

[0008] In the bridge drone disease identification and assessment method based on the YOLOv8 model provided by the present invention, it can also have the following characteristics: wherein, step S1 specifically includes the following steps: S1-1, using a drone high-definition camera to photograph key defects of key components of the highway bridge structure, bridge deck pavement, superstructure and substructure, and collect disease image information; S1-2, manually classifying the drone disease image pictures taken according to longitudinal joints, subsidence, bulges, rutting, potholes, looseness, spalling, oil overflow, repairs and structural cracks; S1-3, using Labelme to frame and annotate the data set, and exporting the annotated images as a label set in yolo format for model training.

[0009] In the bridge drone disease identification and assessment method based on the YOLOv8 model provided by the present invention, it can also have the following characteristics: wherein, step S2 specifically includes the following steps: S2-1, using VS Code and Anaconda to set up a deep learning training environment, build a pytorch deep learning framework, and install CUDA tools; S2-2, obtaining the YOLOv8 deep model basic framework from Ultralystic, and testing the basic performance of the YOLO model based on the deep learning training environment established in S2-1; S2-3, considering that the scale of diseases such as cracks and longitudinal seams is small, in order to improve the detection capability of the scale model of the YOLOv8 model, a 160×160 detection feature map is added to the original three detection heads of multi-scale feature fusion for processing targets of large (32×32), medium (16×16), and small (8×8) scales, for detecting targets larger than 4×4; S2-4, in order to improve the detection accuracy, the CBAM attention mechanism combining channel attention and spatial attention is introduced into the YOLOv8 model.

[0010] In the bridge drone disease identification and assessment method based on the YOLOv8 model provided by the present invention, it can also have the following characteristics: wherein, the specific modification method of step S2-3 is as follows: the original configuration file and model structure of the YOLOv8 model are stored in the YOLOv8.yaml file, firstly, the head to be outputted in the 15th layer in the original Neck is replaced with the Upsample-Concat-C2F module group, which is used for the 160×160 head output of the small target detection branch, and then the CBS-Concat-C2F module group is added behind the 160×160 head output, and the 160×160 convolution is first downsampled to 80×80, and the 80×80 head is outputted in contact with the 15th layer output in the Neck.

[0011] In the bridge drone disease identification and assessment method based on the YOLOv8 model provided by the present invention, it can also have the following characteristics: wherein, step S2-4 is specifically modified as follows: CBAM includes two sub-modules, a channel attention module and a spatial attention module, which perform channel and spatial attention respectively, wherein the channel attention module performs maximum pooling and average pooling on the input feature map in the channel dimension, and then inputs the two pooling results into a fully connected layer, and finally outputs a channel attention weight vector, which can be used to weight each channel of the input feature map to enhance important channel features and suppress unimportant channel features, and the spatial attention module calculates the importance of each pixel by operating the input feature map, and the spatial attention module usually uses global average pooling to obtain the feature vector of each pixel, and then outputs the weight of each pixel through a fully connected layer, and the weight is used to weight each pixel of the input feature map to emphasize the important areas in the image and suppress the unimportant areas, and the channel attention module and the spatial attention module are connected to establish a CBAM module, and the CBAM module is imported into the YOLOv8.yaml module.

[0012] In the bridge drone disease identification and assessment method based on the YOLOv8 model provided by the present invention, it can also have the following characteristics: wherein, step S3 specifically includes the following steps: S3-1, based on a preset data set and a deep model framework for bridge drone disease identification that has been established, the key indicators such as mAP and detection rate of five models, namely YOLOv8n, YOLOv8s, YOLOv8m, YOLOv8l and YOLOv8x, are tested respectively, and a suitable detection model and number of training epochs are selected; S3-2, based on the trained model, the drone detection video stream is classified and detected, and a qualitative evaluation process for the bridge structure is established.

[0013] The method for identifying and assessing bridge defects using drones based on the YOLOv8 model provided by the present invention may also have the following features: wherein the qualitative assessment process of the bridge structure in step S3-2 is specifically as follows: (1) drone image inspection of bridge facilities; (2) bridge defect detection based on the YOLOv8 algorithm; (3) classification of defect detection results; (4) qualitative assessment of bridge defects; and (5) bridge operation and maintenance recommendations.

[0014] In the bridge drone disease identification and assessment method based on the YOLOv8 model provided by the present invention, it can also have the following characteristics: wherein, step S4 specifically includes the following steps: S4-1, using a drone to inspect the bridge structure, and photographing the key components of the bridge deck system, superstructure and substructure in turn; S4-2, writing code to use the YOLOv8 detection model to perform real-time detection on the inspection video data stream, and perform a qualitative assessment of the bridge according to the qualitative assessment algorithm of the bridge structure obtained in step S3.

[0015] Functions and Effects of the Invention

[0016] According to the bridge drone disease identification and assessment method based on the YOLOv8 model involved in the present invention, the present invention uses an intelligent method to perform real-time detection, identification and qualitative analysis of bridge apparent diseases, and by classifying the disease types, improves the adaptability of the labeled image and the model to be trained, and can effectively detect and identify diseases of different scales, effectively improving the detection efficiency and the accuracy of disease identification, and providing strong technical support for the early discovery and timely maintenance of bridge diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a picture of cracks between the pavement and the beam in this embodiment;

[0018] Figure 2 is a picture of disease picture classification in this embodiment;

[0019] Figure 3 is a picture of the image collection and annotation process in this embodiment;

[0020] Figure 4 This is a picture of the basic framework of the YOLOv8 deep model obtained in this embodiment;

[0021] Figure 5 It is a structural diagram of the original configuration file of the YOLOv8 model and the model structure stored in the YOLOv8.yaml file in this embodiment;

[0022] Figure 6 This is a picture of the code modification content in YOLOv8.yaml in this embodiment;

[0023] Figure 7 is a code image for enhancing important channel features and suppressing unimportant channel features in this embodiment;

[0024] Figure 8 This is a code picture of the spatial attention module in this embodiment obtaining the feature vector of each pixel;

[0025] Fig. 9It is a code diagram of connecting the channel attention module and the spatial attention module to establish the CBAM module in this embodiment;

[0026] Fig.10 This is a code diagram for importing the CBAM module into the YOLOv8.yaml module in this embodiment;

[0027] Fig.11 It is a code picture of building a model in this embodiment;

[0028] Fig.12 is a picture of the calculation results of each model in this embodiment;

[0029] Fig.13 is a qualitative evaluation flow chart of the bridge structure in this embodiment;

[0030] Fig.14 is a photo of the inspection process in this embodiment;

[0031] Fig.15 is the real-time detection code of the drone transmission video in this embodiment; and

[0032] Fig.16 3 is a diagram showing the detection results of the pavement portion of the example bridge in this embodiment. DETAILED DESCRIPTION

[0033] In the description of this application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, an electrical connection, or mutual communication; it can be a direct connection, or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0034] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the following embodiments and the accompanying drawings specifically illustrate the bridge drone disease identification and assessment method based on the YOLOv8 model of the present invention.

[0035] The bridge drone disease identification and assessment method based on the YOLOv8 model in this embodiment specifically includes the following steps:

[0036] S1, using drone high-definition cameras to photograph key defects of key components of highway bridge structures, including bridge deck pavement and superstructure, collect defect image information, and use Labelme to annotate the data set into yolo training format to obtain annotated data set.

[0037] S1 includes the following sub-steps:

[0038] S1-1, use the high-definition camera of the drone to take photos of key defects in key components of highway bridge structures, bridge deck pavement, superstructure and substructure, and collect defect image information.

[0039] Figure 1 2 is a picture of the cracks between the pavement and the beam in this embodiment.

[0040] like Figure 1 As shown, the main types of defects include longitudinal cracks, subsidence, bulges, rutting, potholes, looseness, spalling, oil spills, repairs and structural cracks. When the defect type is a small sample defect type, the model is modified to add a large-size detection head.

[0041] Figure 2 It is a picture of disease picture classification in this embodiment.

[0042] S1-2, such as Figure 2 As shown in the figure, the drone images of damage were manually classified according to longitudinal joints, subsidence, bulges, rutting, potholes, looseness, spalling, oil overflow, repairs and structural cracks.

[0043] Figure 3 is a picture of the image collection and annotation process in this embodiment.

[0044] S1-3, such as Figure 3 As shown in the figure, Labelme is used to annotate the dataset, and the annotated images are exported as label sets in yolo format for model training.

[0045] S2, use VS Code to build the basic framework of the YOLOv8 deep model. For small sample disease types, modify the model to add a larger detection head, introduce the CBAM attention mechanism into the YOLOv8 model, thereby improving the detection accuracy and forming a deep model framework suitable for bridge drone disease identification.

[0046] S2 includes the following sub-steps:

[0047] S2-1, use VS Code and Anaconda to set up a deep learning training environment, build the pytorch deep learning framework, and install CUDA tools.

[0048] Figure 4 This is a picture of the basic framework of the YOLOv8 deep model obtained in this embodiment.

[0049] S2-2, such as Figure 4As shown in the figure, the YOLOv8 deep model basic framework is obtained from Ultralystic, and the basic performance of the YOLO model is tested based on the deep learning training environment established in S2-1.

[0050] S2-3, considering that the scale of defects such as cracks and longitudinal seams is small, in order to improve the detection capability of the scale model of the YOLOv8 model, a 160×160 detection feature map is added to the original three detection heads of multi-scale feature fusion for processing targets of large (32×32), medium (16×16), and small (8×8) scales, which is used to detect targets larger than 4×4.

[0051] The specific modification methods are as follows:

[0052] Figure 5 It is a structural diagram of the original configuration file of the YOLOv8 model and the model structure stored in the YOLOv8.yaml file in this embodiment. Figure 6 This is a picture of the code modification content in YOLOv8.yaml in this embodiment.

[0053] like Figure 5-6 As shown in the figure, the original configuration file and model structure of the YOLOv8 model are stored in the YOLOv8.yaml file. First, the head that should be output in the 15th layer in the original Neck is replaced with the Upsample-Concat-C2F module group for the 160×160 head output of the small target detection branch. Then, the CBS-Concat-C2F module group is added after the 160×160 head output, and the 160×160 convolution is downsampled to 80×80, which is contacted with the 15th layer output in Neck to output the 80×80 head.

[0054] S2-4, in order to improve the detection accuracy, the CBAM (Convolutional Block Attention Module) attention mechanism that combines channel attention and spatial attention is introduced into the YOLOv8 model.

[0055] The specific modification methods are as follows:

[0056] Figure 7 It is a code picture that enhances important channel features and suppresses unimportant channel features in this embodiment.

[0057] like Figure 7As shown in the figure, CBAM includes two sub-modules: channel attention module and spatial attention module. The channel attention module is CAM (Channel Attention Module) and the spatial attention module is SAM (Spartial Attention Module), which perform channel and spatial attention respectively. The channel attention module performs maximum pooling and average pooling on the input feature map in the channel dimension, and then inputs the two pooling results into a fully connected layer, and finally outputs a channel attention weight vector. The weight vector can be used to weight each channel of the input feature map to enhance important channel features and suppress unimportant channel features.

[0058] Figure 8 It is a code picture of the spatial attention module in this embodiment to obtain the feature vector of each pixel.

[0059] The spatial attention module calculates the importance of each pixel by operating on the input feature map. The spatial attention module usually uses global average pooling to obtain the feature vector of each pixel, and then outputs the weight of each pixel through a fully connected layer. The weight is used to weight each pixel of the input feature map to emphasize important areas in the image and suppress unimportant areas.

[0060] Fig. 9 This is a code diagram of connecting the channel attention module and the spatial attention module to establish a CBAM module in this embodiment.

[0061] like Fig. 9 As shown, the two are connected to establish a CBAM module.

[0062] Fig.10 This is a code diagram for importing the CBAM module into the YOLOv8.yaml module in this embodiment.

[0063] like Fig.10 As shown, import the CBAM module into the YOLOv8.yaml module.

[0064] S3, based on the labeled dataset and the built deep model framework suitable for bridge drone disease identification, tested the key indicators such as mAP and detection rate of five models, YOLOv8n, YOLOv8s, YOLOv8m, YOLOv8l and YOLOv8x, selected the appropriate detection model and the number of training epochs, classified the detection results of the drone detection video stream based on the trained model, and established a qualitative evaluation algorithm for the bridge structure.

[0065] S3 includes the following sub-steps:

[0066] Fig.11 It is a code picture of building a model in this embodiment. Fig.12 2 is a picture of the calculation results of each model in this embodiment.

[0067] S3-1, such as Figure 11-12 As shown in the figure, based on the set data set and the built deep model framework for bridge drone disease recognition, the key indicators such as mAP and detection rate of five models, YOLOv8n, YOLOv8s, YOLOv8m, YOLOv8l and YOLOv8x, are tested respectively, and the appropriate detection model and number of training epochs are selected.

[0068] S3-2, based on the trained model, the detection results of the drone detection video stream are classified and a qualitative evaluation process of the bridge structure is established.

[0069] Fig.13 It is a qualitative evaluation flow chart of the bridge structure in this embodiment.

[0070] like Fig.13 As shown in Figure 2, the qualitative assessment process of bridge structures is as follows:

[0071] (1) UAV image inspection of bridge facilities.

[0072] (2) Bridge damage detection based on YOLOv8 algorithm.

[0073] (3) Classification of disease detection results.

[0074] (4) Qualitative assessment of bridge defects.

[0075] (5) Bridge operation and maintenance recommendations.

[0076] Table 1

[0077]

[0078]

[0079] As shown in Table 1, it is the basis for qualitative evaluation of honeycomb and crack defects.

[0080] S4, using drones to inspect bridge structures, using detection models to perform real-time detection of inspection video data streams, and performing qualitative evaluations using qualitative evaluation algorithms for bridge structures.

[0081] S4 includes the following sub-steps:

[0082] Fig.14 2 is a photo of the inspection process in this embodiment. Fig.15 It is the real-time detection code of the drone transmission video in this embodiment.

[0083] S4-1, such as Figure 14-15As shown, a drone is used to inspect the bridge structure, and the key components of the bridge deck, superstructure and substructure are photographed in sequence. In this embodiment, a highway in Shanghai is taken as an example, and the bridge structure information is photographed by a drone to transmit real-time data to a computer server.

[0084] S4-2, write code to use the YOLOv8 detection model to perform real-time detection on the inspection video data stream, and perform a qualitative assessment of the bridge according to the qualitative assessment algorithm of the bridge structure obtained in step S3.

[0085] Fig.16 3 is a diagram showing the detection results of the pavement portion of the example bridge in this embodiment.

[0086] like Fig.16 As shown, the main defects of the bridge deck are potholes, longitudinal cracks and transverse cracks. There are 1 unrepaired longitudinal crack and 1 transverse crack respectively, the longitudinal crack scale is 3, and the transverse crack scale is 2, which need to be repaired immediately. The number of unrepaired potholes is 1, and the scale is 2, which also need to be repaired immediately.

[0087] Functions and Effects of the Embodiments

[0088] According to the bridge drone disease identification and assessment method based on the YOLOv8 model involved in the present invention, the present invention uses an intelligent method to perform real-time detection, identification and qualitative analysis of bridge apparent diseases, and by classifying the disease types, improves the adaptability of the labeled image and the model to be trained, and can effectively detect and identify diseases of different scales, effectively improving the detection efficiency and the accuracy of disease identification, and providing strong technical support for the early discovery and timely maintenance of bridge diseases.

[0089] The three detection heads of the present invention that originally used multi-scale feature fusion are used to process targets of different scales, large (32×32), medium (16×16), and small (8×8), and a 160×160 detection feature map is added to detect targets larger than 4×4, which effectively improves the detection capability of the scale model of the YOLOv8 model, so that smaller-scale cracks and longitudinal seams and other defects can also be better detected.

[0090] The present invention introduces the CBAM attention mechanism which combines channel attention and spatial attention into the YOLOv8 model, thereby forming a deep model framework suitable for bridge drone disease identification, which effectively improves the detection accuracy.

[0091] Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A bridge drone disease identification and assessment method based on the YOLOv8 model, characterized in that: The specific steps include: S1, using drone high-definition cameras to take photos of key defects of key components of highway bridge structures, including bridge deck pavement and superstructure, collect defect image information, and use Labelme to annotate the data set into the yolo training format to obtain annotated data set; S2, using VS Code to build the YOLOv8 deep model basic framework. When the type of the disease is a small sample disease type, the model is modified to add a larger detection head and introduce the CBAM attention mechanism into the YOLOv8 model, thereby improving the detection accuracy and forming a deep model framework suitable for bridge drone disease recognition; S3, based on the labeled data set and the built deep model framework suitable for bridge drone disease identification, test the key indicators of mAP and detection rate of five models, namely YOLOv8n, YOLOv8s, YOLOv8m, YOLOv8l and YOLOv8x, respectively, select the appropriate detection model and the number of training epochs, classify the detection results of the drone detection video stream based on the trained model, and establish a qualitative evaluation algorithm for the bridge structure; S4, using a drone to inspect the bridge structure, using a detection model to perform real-time detection on the inspection video data stream, and performing a qualitative evaluation of the bridge structure using a qualitative evaluation algorithm.

2. The bridge drone disease identification and assessment method based on the YOLOv8 model according to claim 1 is characterized in that: in, The defects in step S1 include the following types: longitudinal cracks, subsidence, bulges, rutting, potholes, looseness, spalling, oil overflow, repairs and structural cracks.

3. The bridge drone disease identification and assessment method based on the YOLOv8 model according to claim 2 is characterized in that: in, Step S1 specifically includes the following steps: S1-1, using drone high-definition cameras to photograph key structural components, bridge deck pavement, superstructure and substructure of highway bridges for key defects and collect defect image information; S1-2: Manually classify the drone-photographed images of damage according to longitudinal cracks, subsidence, bulges, rutting, potholes, looseness, spalling, oil spills, repairs, and structural cracks; S1-3, use Labelme to annotate the dataset, and export the annotated images as a label set in yolo format for model training.

4. The bridge drone disease identification and assessment method based on the YOLOv8 model according to claim 1 is characterized in that: in, Step S2 specifically includes the following steps: S2-1, use VS Code and Anaconda to set up a deep learning training environment, build the pytorch deep learning framework, and install CUDA tools; S2-2, obtain the YOLOv8 deep model basic framework from Ultralystic, and test the basic performance of the YOLO model based on the deep learning training environment established in S2-1; S2-3, considering that the scale of defects such as cracks and longitudinal seams is small, in order to improve the detection capability of the scale model of the YOLOv8 model, a 160×160 detection feature map is added to the original three detection heads of multi-scale feature fusion for processing targets of large (32×32), medium (16×16), and small (8×8) scales, for detecting targets larger than 4×4; S2-4, in order to improve the detection accuracy, the CBAM attention mechanism combining channel attention and spatial attention is introduced into the YOLOv8 model.

5. The bridge drone disease identification and assessment method based on the YOLOv8 model according to claim 4 is characterized in that: in, The specific modification method of step S2-3 is as follows: The original configuration file and model structure of the YOLOv8 model are stored in the YOLOv8.yaml file. First, the head that should be output in the 15th layer in the original Neck is replaced with the Upsample-Concat-C2F module group for the 160×160 head output of the small target detection branch. Then, the CBS-Concat-C2F module group is added after the 160×160 head output, and the 160×160 convolution is downsampled to 80×80, and the 80×80 head is output in contact with the 15th layer output in Neck.

6. The bridge drone disease identification and assessment method based on the YOLOv8 model according to claim 4 is characterized in that: in, The specific modification method of step S2-4 is as follows: The CBAM includes two sub-modules, namely a channel attention module and a spatial attention module, which perform attention on channels and space respectively. The channel attention module performs maximum pooling and average pooling on the input feature map in the channel dimension, and then inputs the two pooling results into a fully connected layer, and finally outputs a channel attention weight vector. The weight vector can be used to weight each channel of the input feature map to enhance important channel features and suppress unimportant channel features. The spatial attention module calculates the importance of each pixel by operating the input feature map. The spatial attention module usually uses global average pooling to obtain the feature vector of each pixel, and then outputs the weight of each pixel through a fully connected layer. The weight is used to weight each pixel of the input feature map to emphasize important areas in the image and suppress unimportant areas. The channel attention module and the spatial attention module are connected to establish a CBAM module, and the CBAM module is imported into the YOLOv8.yaml module.

7. The bridge drone disease identification and assessment method based on the YOLOv8 model according to claim 1 is characterized in that: in, Step S3 specifically includes the following steps: S3-1, based on the pre-set data set and the deep model framework for bridge drone disease recognition, test the key indicators such as mAP and detection rate of five models, YOLOv8n, YOLOv8s, YOLOv8m, YOLOv8l and YOLOv8x, and select the appropriate detection model and number of training epochs; S3-2, based on the trained model, the detection results of the drone detection video stream are classified and a qualitative evaluation process of the bridge structure is established.

8. The bridge drone disease identification and assessment method based on the YOLOv8 model according to claim 1 is characterized in that: in, The qualitative evaluation process of the bridge structure in step S3-2 is as follows: (1) UAV image inspection of bridge facilities; (2) Bridge damage detection based on YOLOv8 algorithm; (3) Classification of disease detection results; (4) Qualitative assessment of bridge damage; (5) Bridge operation and maintenance recommendations.

9. The bridge drone disease identification and assessment method based on the YOLOv8 model according to claim 1 is characterized in that: in, Step S4 specifically includes the following steps: S4-1, using a drone to inspect the bridge structure, and photographing key components of the bridge deck system, superstructure and substructure in sequence; S4-2, write code to use the YOLOv8 detection model to perform real-time detection on the inspection video data stream, and perform a qualitative assessment of the bridge according to the qualitative assessment algorithm of the bridge structure obtained in step S3.

Images