Aircraft surface defect detection method based on improved YOLOv8

Abstract

The invention relates to the technical field of virtual reality and simulation, and relates to an improved YOLOv8-based aircraft surface defect detection method, which comprises the following steps of: 1, collecting aircraft surface defect image data, marking the aircraft surface defect image data, and expanding a data set; 2, analyzing the airplane surface defect data set obtained in the step 1 based on a YOLOv8 model, improving the YOLOv8 model, and improving the detection performance of the YOLOv8 model; 3, building an aircraft surface defect detection system by using the YOLOv8 model improved in the step 2 and based on interactive visual editing software; 4, an improved YOLOv8 model is combined with the aircraft surface defect detection system built in the step 3 for defect detection, the problem that precision and real-time detection performance cannot be considered at the same time in existing aircraft surface defect detection is solved, and high-precision rapid detection of the aircraft surface is achieved.

Description

Technical Field

[0001] The present invention relates to the field of virtual reality and simulation technology, and in particular to an aircraft surface defect detection method based on improved YOLOv8. Background Art

[0002] Aircraft surface defect detection is a key link in aviation manufacturing and maintenance. The safety of aircraft is directly related to the life safety of passengers and crew members. Therefore, timely and accurate detection of damage and defects on the aircraft surface is crucial. Common defects include cracks, scratches, dents and coating peeling. If these defects are not discovered and handled in time, they may lead to reduced structural strength and even cause serious accidents. Traditional manual inspection methods still occupy an important position in the aviation field, but this method is not only time-consuming, but also easily affected by human factors, resulting in inconsistency and subjectivity in the inspection results. In addition, with the expansion of the scale of aircraft manufacturing and maintenance, the requirements for inspection efficiency and accuracy are increasing, and traditional methods are difficult to meet the needs of efficient and accurate inspection.

[0003] In recent years, intelligent detection methods based on computer vision and deep learning have been gradually applied to the aviation field. Intelligent detection not only improves the detection speed, but also effectively reduces human misjudgment. However, existing deep learning detection methods still face many challenges in practical applications, including detection accuracy, real-time requirements, and adaptability in different environments and scenarios. These problems limit the widespread application of intelligent detection technology in the aviation industry. As a single-stage target detection algorithm, the YOLO (You Only Look Once) series of algorithms are widely used in defect detection tasks due to their fast detection speed and high accuracy. YOLOv8 is the latest version of the series. By using improved network structure and training methods, the feature extraction capability and multi-scale detection effect are significantly enhanced. The model is particularly outstanding in processing small-sized objects and complex scenes. The flexibility and scalability of YOLOv8 make it suitable for different types of aircraft surface defect detection tasks, and the network structure can be improved to cope with various environmental conditions and defect types. The YOLOv8 model provides a new solution for defect detection in the field of aviation manufacturing and maintenance, and has important application prospects and practical significance. Summary of the invention

[0004] The purpose of the invention is to provide an aircraft surface defect detection method based on improved YOLOv8, so as to solve the problem that the current aircraft surface defect detection cannot take into account both accuracy and real-time detection performance, and realize high-precision and rapid detection of the aircraft surface.

[0005] The technical solution of the present invention:

[0006] The aircraft surface defect detection method based on improved YOLOv8 includes the following steps:

[0007] Step 1: Collect and annotate aircraft surface defect image data to expand the data set;

[0008] Step 2: Based on the YOLOv8 model, analyze the aircraft surface defect dataset obtained in step 1, improve the YOLOv8 model, and improve the detection performance of the YOLOv8 model;

[0009] Step 3: Use the improved YOLOv8 model from step 2 to build an aircraft surface defect detection system based on interactive visual editing software;

[0010] Step 4: Use the improved YOLOv8 model combined with the aircraft surface defect detection system built in step 3 to perform defect detection.

[0011] Furthermore, in step 1, the data set is expanded using the Mosaic data enhancement method.

[0012] Further, the step 1 comprises the following steps:

[0013] Step 1.1, use LabelImg to label the data set and get the input image:

[0014] Use the LabelImg tool to annotate the collected aircraft surface defect images and generate standardized annotation files. The annotation files contain the defect location and category information in each image for subsequent model training.

[0015] Step 1.2, segment the input image into multiple sub-images;

[0016] Step 1.3, adjusting the annotation box of each segmented image so that each segmented image has accurate annotation information and ensuring that the annotation box matches the segmented image;

[0017] Step 1.4, stitch the segmented images into a new image in units of four to form a Mosaic enhanced image;

[0018] Step 1.5, transform the color gamut of the new image, adjust the hue, saturation and brightness to simulate the image effects under different lighting conditions;

[0019] Step 1.6, merge the annotation boxes to obtain a new image and the corresponding annotation boxes, that is, to obtain a data set after data enhancement and expansion;

[0020] Step 1.7, divide the expanded data set into training set, validation set, and test set with a division ratio of 7:2:1.

[0021] Further, the step 2 specifically includes the following steps:

[0022] Step 2.1, replace the C2f module in YOLOv8 Backbone with the attention mechanism combining ECA and SE modules to obtain the YOLOv8 ECA-SE model; the ECA module can calculate the importance of each channel through convolution kernel operation, and the SE module further enhances the channel weight allocation ability through global average pooling;

[0023] The convolution kernel size of the ECA module is expressed as:

[0024]

[0025] Among them, C is the number of channels, k is the size of the convolution kernel, and Formula 1 can dynamically adjust the receptive field range of the attention mechanism according to the number of channels;

[0026] Step 2.2, based on the YOLOv8 ECA-SE model, add an enhanced feature pyramid network to achieve multi-scale feature fusion and obtain the YOLOv8 ECA-SE EFPN model;

[0027] In the enhanced feature pyramid network, features of different scales are fused through top-down transmission, and the formula is expressed as:

[0028] F′ i =F i +Upsample(F( i+1 )) (2)

[0029] Among them, F i is the i-th layer feature, Upsample(F i+1 ) is the high-level feature transmitted after upsampling; through multi-scale fusion, the perception ability of small defects is improved;

[0030] Step 2.3, replace the bounding box loss function of the YOLOv8 ECA-SE EFPN model with DIoU, and propose the YOLOv8ECA-SE EFPN DIoU aircraft surface defect detection model, that is, the improved YOLOv8 model;

[0031] DIoU loss takes into account the overlapping area of ​​the bounding box and introduces the distance factor between the center points. Its loss function is expressed as:

[0032]

[0033] Among them, ρ is the distance between the center point of the predicted box and the real box, b is the predicted box, and b gtis the real box, c is the diagonal length of the bounding box, and through the introduction of DIoU, the regression effect of the model on the bounding box has been significantly improved. IoU is the intersection over union ratio.

[0034] Furthermore, step 3 specifically uses the improved YOLOv8 model in step 2 and builds an aircraft surface defect detection system based on PyQt5 software.

[0035] Further, step 3 specifically includes the following steps:

[0036] Step 3.1, build the login and registration interface through Qt Designer software;

[0037] Step 3.2, build the defect detection interface through Qt Designer software;

[0038] Step 3.3, build the test result display interface through Qt Designer software and add the defect classification module;

[0039] Step 3.4, build the result saving interface through Qt Designer software;

[0040] Step 3.5, convert the UI file of the interface built in steps 3.1 to 3.4 into a Python file through PYUIC to obtain the aircraft surface defect detection system.

[0041] Further, step 4 specifically includes the following steps:

[0042] Step 4.1, set the YOLOv8 ECA-SE EFPN DIoU model hyperparameters;

[0043] Hyperparameters include learning rate, batch size, and weight decay. For the YOLOv8 ECA-SE EFPN DIoU model, the initial value of the learning rate α is set to 0.001 and the weight decay coefficient λ is set to 0.0005 to ensure rapid convergence of the model.

[0044] Step 4.2, train the YOLOv8 ECA-SE EFPN DIoU model according to the data set divided in step 1, and save the model weights;

[0045] Step 4.3, adding the model weight to the aircraft surface defect detection system built in step 3;

[0046] Step 4.4, run the Python file generated in step 3 to open the aircraft surface defect detection system;

[0047] Step 4.5: Input the image to be inspected into the aircraft surface defect inspection system for offline defect inspection, or turn on the camera switch for real-time defect inspection.

[0048] Furthermore, the real-time defect detection process in step 4.5 is as follows: after the detection system is started, the user turns on the camera through the system interface for real-time defect detection, and the improved YOLOv8 model will automatically identify the aircraft surface defects in the input image and output the detection results including defect category, location and confidence.

[0049] Beneficial effects of the present invention:

[0050] This paper proposes an aircraft surface defect detection algorithm based on an improved YOLOv8 model. First, the C2f module in Backbone is replaced by a hybrid spatial-channel attention mechanism to improve the multi-scale feature extraction capability of the model; then, the problem of detecting tiny defects is solved by introducing an enhanced feature pyramid network; finally, the bounding box regression is optimized by the Dynamic IoU loss function to improve the detection precision and accuracy of aircraft surface defects.

[0051] Through the detection system built by PyQt5 software and combined with the improved model, an efficient real-time aircraft surface defect detection method is proposed. Through the automated detection method, manpower input is reduced and the defect detection efficiency is improved. By setting the detection time threshold, real-time monitoring of the aircraft surface state is achieved, making the detection process more controllable and operational. The present invention helps to improve the detection efficiency and operation process, and reduce the errors and inefficiencies caused by manual detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 : Flowchart of the present invention.

[0053] Figure 2 : Network structure diagram of the present invention.

[0054] Figure 3 : The operating system interface of the present invention.

[0055] Figure 4 : Detection result schematic diagram of the present invention. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] One embodiment of the present invention is to provide an aircraft surface defect detection method based on improved YOLOv8, comprising the following steps:

[0058] Step 1: Collect and annotate aircraft surface defect image data to expand the data set;

[0059] In the step 1, the data set is expanded using the Mosaic data enhancement method; step 1 includes the following steps:

[0060] Step 1.1, use LabelImg to label the data set and get the input image:

[0061] Use the LabelImg tool to annotate the collected aircraft surface defect images and generate standardized annotation files. The annotation files contain the defect location and category information in each image for subsequent model training.

[0062] Step 1.2, segment the input image into multiple sub-images;

[0063] Step 1.3, adjusting the annotation box of each segmented image so that each segmented image has accurate annotation information and ensuring that the annotation box matches the segmented image;

[0064] Step 1.4, stitch the segmented images into a new image in units of four to form a Mosaic enhanced image;

[0065] Step 1.5, transform the color gamut of the new image, adjust the hue, saturation and brightness to simulate the image effects under different lighting conditions;

[0066] Step 1.6, merge the annotation boxes to obtain a new image and the corresponding annotation boxes, that is, to obtain a data set after data enhancement and expansion;

[0067] Step 1.7, divide the expanded data set into training set, validation set, and test set with a division ratio of 7:2:1.

[0068] Step 2: Based on the YOLOv8 model, analyze the aircraft surface defect dataset obtained in step 1, improve the YOLOv8 model, and improve the detection performance of the YOLOv8 model;

[0069] Step 2 specifically includes the following steps:

[0070] Step 2.1, replace the C2f module in YOLOv8 Backbone with the attention mechanism combining ECA and SE modules to obtain the YOLOv8 ECA-SE model; the ECA module can calculate the importance of each channel through convolution kernel operation, and the SE module further enhances the channel weight allocation ability through global average pooling;

[0071] The convolution kernel size of the ECA module is expressed as:

[0072]

[0073] Among them, C is the number of channels, k is the size of the convolution kernel, and Formula 1 can dynamically adjust the receptive field range of the attention mechanism according to the number of channels;

[0074] Step 2.2, based on the YOLOv8 ECA-SE model, add an enhanced feature pyramid network to achieve multi-scale feature fusion and obtain the YOLOv8 ECA-SE EFPN model;

[0075] In the enhanced feature pyramid network, features of different scales are fused through top-down transmission, and the formula is expressed as:

[0076] F i ′=F i +Upsample(F( i+1 )) (2)

[0077] Among them, F i is the i-th layer feature, Upsample(F i+1 ) is the high-level feature transmitted after upsampling; through multi-scale fusion, the perception ability of small defects is improved;

[0078] Step 2.3, replace the bounding box loss function of the YOLOv8 ECA-SE EFPN model with DIoU, and propose the YOLOv8ECA-SE EFPN DIoU aircraft surface defect detection model, that is, the improved YOLOv8 model;

[0079] DIoU loss takes into account the overlapping area of ​​the bounding box and introduces the distance factor between the center points. Its loss function is expressed as:

[0080]

[0081] Among them, ρ is the distance between the center point of the predicted box and the real box, b is the predicted box, and b gt is the real box, c is the diagonal length of the bounding box, and through the introduction of DIoU, the regression effect of the model on the bounding box has been significantly improved. IoU is the intersection over union ratio.

[0082] Step 3: Use the improved YOLOv8 model in step 2 to build an aircraft surface defect detection system based on interactive visual editing software; Step 3 specifically uses the improved YOLOv8 model in step 2 to build an aircraft surface defect detection system based on PyQt5 software, including the following steps:

[0083] Step 3.1, build the login and registration interface through Qt Designer software;

[0084] Step 3.2, build the defect detection interface through Qt Designer software;

[0085] Step 3.3, build the test result display interface through Qt Designer software and add the defect classification module;

[0086] Step 3.4, build the result saving interface through Qt Designer software;

[0087] Step 3.5, convert the UI file of the interface built in steps 3.1 to 3.4 into a Python file through PYUIC to obtain the aircraft surface defect detection system.

[0088] Step 4: Use the improved YOLOv8 model combined with the aircraft surface defect detection system built in step 3 to perform defect detection, which specifically includes the following steps:

[0089] Step 4.1, set the YOLOv8 ECA-SE EFPN DIoU model hyperparameters;

[0090] Hyperparameters include learning rate, batch size, and weight decay. For the YOLOv8 ECA-SE EFPN DIoU model, the initial value of the learning rate α is set to 0.001 and the weight decay coefficient λ is set to 0.0005 to ensure rapid convergence of the model.

[0091] Step 4.2, train the YOLOv8 ECA-SE EFPN DIoU model according to the data set divided in step 1, and save the model weights;

[0092] Step 4.3, adding the model weight to the aircraft surface defect detection system built in step 3;

[0093] Step 4.4, run the Python file generated in step 3 to open the aircraft surface defect detection system;

[0094] Step 4.5, input the image to be detected in the aircraft surface defect detection system for offline defect detection, or turn on the camera switch for real-time defect detection. The real-time defect detection process is: after the detection system is started, the user turns on the camera through the system interface for real-time defect detection. The improved YOLOv8 model will automatically identify the aircraft surface defects in the input image and output the detection results including defect category, location and confidence.

[0095] Another embodiment of the present invention is to provide an aircraft surface defect detection method based on improved YOLOv8. The aircraft surface defect detection method of the present invention first requires a high-quality aircraft surface defect image dataset. According to the technical solution of step 1, the specific steps are as follows:

[0096] 1. Data collection and annotation

[0097] Use the LabelImg tool to annotate the collected aircraft surface defect images and generate standardized annotation files (such as XML format), which contain the defect location and category information in each image. These annotation data will be used for subsequent model training.

[0098] 2. Image segmentation and Mosaic data enhancement

[0099] Each original image is segmented into multiple sub-images, and the corresponding annotation boxes are adjusted so that each segmented image has accurate annotation information. Then, the four segmented images are spliced ​​into a new image to form a Mosaic enhanced image. Mosaic data enhancement can greatly enrich the diversity of data and improve the generalization ability of the model without increasing the workload of annotation.

[0100] 3. Color gamut conversion and data expansion

[0101] The newly generated images are transformed into a color space, and the hue, saturation, and brightness are adjusted to simulate the image effects under different lighting conditions. This processing can enhance the model's adaptability to complex environments. After data enhancement, a new data set is finally formed and divided into a training set, a validation set, and a test set in a ratio of 8:1:1.

[0102] Example 2: Improved YOLOv8 model

[0103] In the process of improving the YOLOv8 model, according to the technical solution of step 2, the specific steps are as follows:

[0104] 1. Introducing the hybrid space-channel attention mechanism

[0105] In order to improve the feature extraction capability of the model, the present invention uses an attention mechanism combining ECA and SE modules to replace the C2f module in YOLOv8 Backbone. The ECA module can calculate the importance of each channel through convolution kernel operations, while the SE module further enhances the weight allocation capability of the channel through global average pooling. Combining these two mechanisms, the YOLOv8 ECA-SE model is formed.

[0106] In this process, the convolution kernel size of the ECA module can be expressed as:

[0107]

[0108] Among them, C is the number of channels and k is the size of the convolution kernel. This formula can dynamically adjust the receptive field range of the attention mechanism according to the number of channels.

[0109] 2. Enhanced Feature Pyramid Network (EFPN)

[0110] On the basis of the YOLOv8 ECA-SE model, the EFPN network is further added for multi-scale feature fusion. EFPN can effectively combine high-level semantic information with low-level detail information, and is particularly suitable for the detection of small-scale defects. In the feature pyramid structure, features of different scales are fused through a top-down transmission method, and the formula is expressed as:

[0111] F i ′=F i +Upsample(F (i+1) ) (2)

[0112] Among them, F i is the i-th layer feature, Upsample(F i+1 ) is the high-level feature transmitted after upsampling. Through this multi-scale fusion, the model's ability to perceive small defects is improved.

[0113] 3. Bounding box loss function replaced by DIoU

[0114] In order to further improve the bounding box regression accuracy of the model, the present invention uses the DIoU loss function to replace the original GIoU loss function. The DIoU loss not only considers the overlapping area of ​​the bounding box, but also introduces the distance factor between the center points. Its loss function can be expressed as:

[0115]

[0116] Among them, ρ is the distance between the center point of the predicted box and the real box, b is the predicted box, and b gt is the true box, and c is the diagonal length of the bounding box. By introducing DIoU, the regression effect of the model on the bounding box is significantly improved.

[0117] Example 3: Construction of detection system based on PyQt5

[0118] The present invention uses PyQt5 software to build a defect detection system, and the specific steps are as follows:

[0119] 1. Interface design

[0120] The login interface, defect detection interface, detection result display interface and result saving interface are designed by Qt Designer. The interface design should be concise and easy to operate, and can support users to upload images of the aircraft surface or obtain images in real time through the camera.

[0121] 2. Model integration and system operation

[0122] The UI file designed by Qt Designer is converted into a Python file using the PYUIC tool, and the improved YOLOv8 ECA-SE EFPN DIoU model mentioned above is embedded into the system. During the operation of the system, users can upload pictures or capture real-time images of the aircraft surface through the graphical interface and start the defect detection process.

[0123] Example 4: Detection system operation

[0124] 1. Hyperparameter settings

[0125] Before model training, hyperparameters need to be set reasonably, including learning rate, batch size, and weight decay. For the improved YOLOv8 ECA-SE EFPN DIoU model, the initial value of the learning rate α is set to 0.001 and the weight decay coefficient λ is set to 0.0005 to ensure rapid convergence of the model.

[0126] 2. Defect detection process

[0127] After the detection system is started, the user can open the camera through the system interface for real-time defect detection. The model will automatically identify the aircraft surface defects in the input image and output the detection results including defect category, location and confidence. Through the multi-scale feature fusion mechanism of EFPN, the system can accurately detect small-sized defects with good real-time performance.

[0128] The above is only a specific embodiment of the present invention, and the present invention is described in detail. The unexplained part is a conventional technology. However, the protection scope of the present invention is not limited to this. Any changes or substitutions that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. The aircraft surface defect detection method based on improved YOLOv8 is characterized by: The following steps are involved: Step 1: Collect and annotate aircraft surface defect image data to expand the data set; Step 2: Based on the YOLOv8 model, analyze the aircraft surface defect dataset obtained in step 1, improve the YOLOv8 model, and improve the detection performance of the YOLOv8 model; Step 3: Use the improved YOLOv8 model from step 2 to build an aircraft surface defect detection system based on interactive visual editing software; Step 4: Use the improved YOLOv8 model combined with the aircraft surface defect detection system built in step 3 to perform defect detection.

2. The aircraft surface defect detection method based on improved YOLOv8 according to claim 1, characterized in that: In step 1, the data set is expanded using the Mosaic data enhancement method.

3. The aircraft surface defect detection method based on improved YOLOv8 according to claim 2, characterized in that: The step 1 comprises the following steps: Step 1.1, use LabelImg to label the data set and get the input image: Use the LabelImg tool to annotate the collected aircraft surface defect images and generate standardized annotation files. The annotation files contain the defect location and category information in each image for subsequent model training. Step 1.2, segment the input image into multiple sub-images; Step 1.3, adjusting the annotation box of each segmented image so that each segmented image has accurate annotation information and ensuring that the annotation box matches the segmented image; Step 1.4, stitch the segmented images into a new image in units of four to form a Mosaic enhanced image; Step 1.5, transform the color gamut of the new image, adjust the hue, saturation and brightness to simulate the image effects under different lighting conditions; Step 1.6, merge the annotation boxes to obtain a new image and the corresponding annotation boxes, that is, to obtain a data set after data enhancement and expansion; Step 1.7, divide the expanded data set into training set, validation set, and test set with a division ratio of 7:2:

1.

4. The aircraft surface defect detection method based on improved YOLOv8 according to claim 1, characterized in that: The step 2 specifically includes the following steps: Step 2.1, replace the C2f module in YOLOv8 Backbone with the attention mechanism combining ECA and SE modules to obtain the YOLOv8 ECA-SE model; the ECA module can calculate the importance of each channel through convolution kernel operation, and the SE module further enhances the channel weight allocation ability through global average pooling; The convolution kernel size of the ECA module is expressed as: Among them, C is the number of channels, k is the size of the convolution kernel, and Formula 1 can dynamically adjust the receptive field range of the attention mechanism according to the number of channels; Step 2.2, based on the YOLOv8 ECA-SE model, an enhanced feature pyramid network is added to realize multi-scale feature fusion to obtain the YOLOv8 ECA-SE EFPN model; In the enhanced feature pyramid network, features of different scales are fused through top-down transmission, and the formula is expressed as: F′ i =F i +Upsample(F (i+1) ) (2) Among them, F i is the i-th layer feature, Upsample(F i+1 ) is the high-level feature transmitted after upsampling; through multi-scale fusion, the perception ability of small defects is improved; Step 2.3, replace the bounding box loss function of the YOLOv8 ECA-SE EFPN model with DIoU, and propose the YOLOv8ECA-SE EFPN DIoU aircraft surface defect detection model, that is, the improved YOLOv8 model; DIoU loss takes into account the overlapping area of ​​the bounding box and introduces the distance factor between the center points. Its loss function is expressed as: Among them, ρ is the distance between the center point of the predicted box and the real box, b is the predicted box, and b gt is the real box, c is the diagonal length of the bounding box, and through the introduction of DIoU, the regression effect of the model on the bounding box has been significantly improved. IoU is the intersection over union ratio.

5. The aircraft surface defect detection method based on improved YOLOv8 according to claim 1, characterized in that: Step 3 specifically uses the improved YOLOv8 model in step 2 and builds an aircraft surface defect detection system based on PyQt5 software.

6. The aircraft surface defect detection method based on improved YOLOv8 according to claim 5 is characterized in that: Step 3 specifically includes the following steps: Step 3.1, build the login and registration interface through Qt Designer software; Step 3.2, build the defect detection interface through Qt Designer software; Step 3.3, build the test result display interface through Qt Designer software and add the defect classification module; Step 3.4, build the result saving interface through Qt Designer software; Step 3.5, convert the UI file of the interface built in steps 3.1 to 3.4 into a Python file through PYUIC to obtain the aircraft surface defect detection system.

7. The aircraft surface defect detection method based on improved YOLOv8 according to claim 6 is characterized in that: Step 4 specifically includes the following steps: Step 4.1, set the YOLOv8 ECA-SE EFPN DIoU model hyperparameters; Hyperparameters include learning rate, batch size, and weight decay. For the YOLOv8 ECA-SE EFPN DIoU model, the initial value of the learning rate α is set to 0.001 and the weight decay coefficient λ is set to 0.0005 to ensure rapid convergence of the model. Step 4.2, train the YOLOv8 ECA-SE EFPN DIoU model according to the data set divided in step 1, and save the model weights; Step 4.3, adding the model weight to the aircraft surface defect detection system built in step 3; Step 4.4, run the Python file generated in step 3 to open the aircraft surface defect detection system; Step 4.5: Input the image to be inspected into the aircraft surface defect inspection system for offline defect inspection, or turn on the camera switch for real-time defect inspection.

8. The aircraft surface defect detection method based on improved YOLOv8 according to claim 7, characterized in that: The real-time defect detection process in step 4.5 is as follows: after the detection system is started, the user opens the camera through the system interface for real-time defect detection. The improved YOLOv8 model will automatically identify the aircraft surface defects in the input image and output the detection results including defect category, location and confidence.

Images