An aircraft skin damage detection method, system and storage medium

Abstract

The application discloses an aircraft skin damage detection method and system and a storage medium, and comprises the following steps: acquiring aircraft skin damage image data and preprocessing; converting the preprocessed data into YOLO format data as a training data set; replacing the hyperbolic tangent function in the additive attention mechanism with a Sigmoid activation function, and replacing the input of the additive attention mechanism with two feature layers of the same size and different semantic information to obtain an improved additive attention mechanism; replacing the fusion mode of the neck feature fusion layer in the YOLO model with the improved additive attention mechanism, inputting the training data set into the improved YOLO model for training to obtain an optimal weight model; inputting the to-be-detected aircraft skin damage image data into the model for detection to obtain an aircraft skin detection result; and effectively solving the problem of model parameter redundancy generated when processing aircraft skin damage detection.

Description

Technical Field

[0001] This invention relates to the field of deep learning and target detection technology, and in particular to a method, system and storage medium for detecting damage to aircraft skin. Background Technology

[0002] Scratches, cracks, corrosion, deformation, and rivet damage severely reduce the strength of aircraft skin structures, significantly impacting flight safety. Therefore, aircraft skin damage detection is a crucial aspect of large aircraft maintenance. Aircraft skin damage datasets often feature images with simple backgrounds, resulting in sparser correlations between the foreground and background of damaged targets compared to publicly available datasets like VOC. Currently, most existing target detection models are designed for complex background data, neglecting the potential for parameter redundancy issues that can arise from applications with simple backgrounds and low-complexity data.

[0003] The neck feature pyramid of existing YOLO series object detection models uses a concat feature fusion method. While stacking feature layer channels does not lead to excessive information loss, it can cause redundancy in model parameters for low-complexity image data such as aircraft skin damage. Summary of the Invention

[0004] This invention discloses a method, system, and storage medium for detecting aircraft skin damage, which can effectively solve the problem of redundant model parameters when processing aircraft skin damage detection.

[0005] To address the aforementioned technical problems, embodiments of the present invention provide a method for detecting damage to aircraft skin, the method comprising: Acquire aircraft skin damage image data and preprocess the aircraft skin damage image data; The preprocessed aircraft skin damage image data is converted to YOLO format data, and the YOLO format data is used as the training dataset. The improved additive attention mechanism is obtained by replacing the hyperbolic tangent function in the additive attention mechanism with the sigmoid activation function and replacing the input of the additive attention mechanism with two feature layers of the same size but different semantic information. By replacing the concat feature fusion method in the neck feature fusion layer of the YOLO model with an improved additive attention mechanism, an improved YOLO model is obtained. The training dataset is input into the improved YOLO model for training to obtain the optimal weight model; The optimal weight model is evaluated to obtain the evaluation test results; If the evaluation test results meet the preset project application requirements, the aircraft skin damage image data to be tested is input into the optimal weight model to detect aircraft skin damage and obtain the aircraft skin detection results.

[0006] Preferably, before obtaining simulation test results by conducting simulation experiments on the optimal weight model, the method further includes: The feature layer is then subjected to visual analysis.

[0007] Preferably, the visualization analysis of the feature layer includes: The visualized sample data is input into the YOLO model and the optimal weight model respectively for forward propagation calculation, and the output results of the feature layer are recorded respectively; wherein the output results include activation maps; The activation maps generated by the YOLO model and the optimal weight model are visualized respectively. The analysis results are generated by comparing and analyzing activation maps, calculating difference images, or drawing histograms.

[0008] Preferably, the improved additive attention mechanism includes at least two feature layers of the same size but different semantic information, a sigmoid activation function, and a Mut operation module; The feature layer is used to extract different semantic information from the input image; The sigmoid activation function is used to map the attention result to the [0,1] interval; The Mut operation module is used to calculate the similarity between the feature layers and to calculate the interaction weights between the feature layers based on the similarity.

[0009] Preferably, the formula for the additive attention mechanism is:

[0010] In the formula, q represents the query vector, k represents the feature vector, and W v Let W represent a weight vector. q W represents the linear transformation matrix of the query vector. k Let represent the linear transformation matrix of the eigenvectors, tanh represent the hyperbolic tangent function, and T represent the linear transformation matrix; The formula for the improved additive attention mechanism is:

[0011] In the formula, high_layer represents high-level semantic information, low_layer represents low-level semantic information, and sigmoid represents the sigmoid activation function.

[0012] Preferably, the step of evaluating the optimal weight model to obtain the evaluation test results includes: By inputting a preset test dataset into the optimal weight model, the model is tested and the model test results are generated. The model test results are compared and analyzed with the true label values ​​of the test data to obtain the model test error value; When the model test error value is less than the preset error value standard, the evaluation test result of the optimal weight model meets the preset project application requirements.

[0013] Preferably, the preprocessing of the aircraft skin damage image data includes: The aircraft skin damage image data was filtered to obtain aircraft skin damage image data in which the background of the images was entirely the aircraft fuselage. Then, remove aircraft skin damage image data from those where the number of targets exceeds the preset target number standard; Then, the aircraft skin damage image data is annotated using a preset data annotation tool.

[0014] This invention discloses a method for detecting aircraft skin damage, comprising: acquiring and preprocessing aircraft skin damage image data; converting the preprocessed data to YOLO format data as a training dataset; replacing the hyperbolic tangent function in the additive attention mechanism with an activation function, and replacing the input of the additive attention mechanism with two feature layers of the same size but different semantic information, to obtain an improved additive attention mechanism; replacing the fusion method of the neck feature fusion layer in the YOLO model with the improved additive attention mechanism, and inputting the training dataset into the improved YOLO model for training to obtain the optimal weight model; inputting the aircraft skin damage image data to be detected into the model for detection, and obtaining the aircraft skin detection result. By combining the additive attention mechanism in the field of natural language processing with the neck feature fusion part of YOLO, an additive feature fusion attention mechanism is proposed, which makes the model pay more attention to the target part to be detected, while compressing the neck feature layer channels of the model, which can meet the requirements of high accuracy and lightweight, so as to better detect damage and fill the gap in related technologies. Therefore, it can effectively solve the problem of redundant model parameters in processing aircraft skin damage detection.

[0015] To address the aforementioned technical problems, embodiments of the present invention also provide an aircraft skin damage detection system, the system comprising: The data processing module is used to acquire aircraft skin damage image data and preprocess the aircraft skin damage image data. The format conversion module is used to convert the preprocessed aircraft skin damage image data into YOLO format data, and use the YOLO format data as a training dataset. The mechanism optimization module is used to replace the hyperbolic tangent function in the additive attention mechanism with the sigmoid activation function, and to replace the input of the additive attention mechanism with two feature layers of the same size but different semantic information, so as to obtain the improved additive attention mechanism. The model optimization module is used to obtain an improved YOLO model by replacing the concat feature fusion method in the neck feature fusion layer of the YOLO model with an improved additive attention mechanism. The model training module is used to input the training dataset into the improved YOLO model for training to obtain the optimal weight model. The model testing module is used to evaluate the optimal weight model and obtain the evaluation test results. The aircraft skin damage detection module is used to input the image data of the aircraft skin damage to be tested into the optimal weight model to detect the aircraft skin damage if the evaluation test results meet the preset project application requirements, and obtain the aircraft skin damage detection results.

[0016] Preferably, the system further includes a visualization module; The visualization module is used to perform visual analysis on the feature layer.

[0017] Preferably, the visualization analysis of the feature layer includes: The visualized sample data is input into the YOLO model and the optimal weight model respectively for forward propagation calculation, and the output results of the feature layer are recorded respectively; wherein the output results include activation maps; The activation maps generated by the YOLO model and the optimal weight model are visualized respectively. The analysis results are generated by comparing and analyzing activation maps, calculating difference images, or drawing histograms.

[0018] Preferably, the improved additive attention mechanism includes at least two feature layers of the same size but different semantic information, a sigmoid activation function, and a Mut operation module; The feature layer is used to extract different semantic information from the input image; The sigmoid activation function is used to map the attention result to the [0,1] interval; The Mut operation module is used to calculate the similarity between the feature layers and to calculate the interaction weights between the feature layers based on the similarity.

[0019] Preferably, the formula for the additive attention mechanism is:

[0020] In the formula, q represents the query vector, k represents the feature vector, and W v Let W represent a weight vector. q W represents the linear transformation matrix of the query vector. k Let represent the linear transformation matrix of the eigenvectors, tanh represent the hyperbolic tangent function, and T represent the linear transformation matrix; The formula for the improved additive attention mechanism is:

[0021] In the formula, high_layer represents high-level semantic information, low_layer represents low-level semantic information, and sigmoid represents the sigmoid activation function.

[0022] Preferably, the step of evaluating the optimal weight model to obtain the evaluation test results includes: By inputting a preset test dataset into the optimal weight model, the model is tested and the model test results are generated. The model test results are compared and analyzed with the true label values ​​of the test data to obtain the model test error value; When the model test error value is less than the preset error value standard, the evaluation test result of the optimal weight model meets the preset project application requirements.

[0023] Preferably, the preprocessing of the aircraft skin damage image data includes: The aircraft skin damage image data was filtered to obtain aircraft skin damage image data in which the background of the images was entirely the aircraft fuselage. Then, remove aircraft skin damage image data from those where the number of targets exceeds the preset target number standard; Then, the aircraft skin damage image data is annotated using a preset data annotation tool.

[0024] The present invention also discloses a computer-readable storage medium comprising a stored computer program; wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform any of the above-described aircraft skin damage detection methods.

[0025] This invention discloses a method, system, and storage medium for aircraft skin damage detection, comprising: acquiring aircraft skin damage image data and preprocessing the aircraft skin damage image data; converting the preprocessed aircraft skin damage image data to YOLO format data, and using the YOLO format data as a training dataset; replacing the hyperbolic tangent function in the additive attention mechanism with a sigmoid activation function, and replacing the input of the additive attention mechanism with two feature layers of the same size but different semantic information, to obtain an improved additive attention mechanism; replacing the concat feature fusion method of the neck feature fusion layer in the YOLO model with the improved additive attention mechanism, to obtain an improved YOLO model; inputting the training dataset into the improved YOLO model for training to obtain an optimal weight model; evaluating the optimal weight model to obtain evaluation test results; if the evaluation test results meet the preset project application requirements, inputting the aircraft skin damage image data to be tested into the optimal weight model for aircraft skin damage detection to obtain aircraft skin detection results. This effectively solves the problem of redundant model parameters in aircraft skin damage detection. Attached Figure Description

[0026] Figure 1 This is a schematic flowchart of an aircraft skin damage detection method provided in an embodiment of the present invention.

[0027] Figure 2 This is a structural block diagram of an aircraft skin damage detection system provided in an embodiment of the present invention.

[0028] Figure 3 This is a structural diagram of the improved additive feature fusion attention mechanism provided in the embodiments of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] This invention provides a method for detecting damage to aircraft skin, see [link to relevant documentation]. Figure 1 This is a flowchart illustrating an aircraft skin damage detection method provided in an embodiment of the present invention, including steps S10 to S16: S10. Acquire aircraft skin damage image data and preprocess the aircraft skin damage image data. S11. Convert the preprocessed aircraft skin damage image data to YOLO format data, and use the YOLO format data as a training dataset. S12. Replace the hyperbolic tangent function in the additive attention mechanism with the sigmoid activation function, and replace the input of the additive attention mechanism with two feature layers of the same size but different semantic information to obtain the improved additive attention mechanism. S13. By replacing the concat feature fusion method in the neck feature fusion layer of the YOLO model with the improved additive attention mechanism, the improved YOLO model is obtained. S14. Input the training dataset into the improved YOLO model for training to obtain the optimal weight model; S15. Perform model evaluation on the optimal weight model to obtain the evaluation test results; S16. If the evaluation test results meet the preset project application requirements, the aircraft skin damage image data to be tested is input into the optimal weight model to detect aircraft skin damage and obtain the aircraft skin detection results.

[0031] This invention discloses a method for detecting aircraft skin damage, comprising: acquiring and preprocessing aircraft skin damage image data; converting the preprocessed data to YOLO format data as a training dataset; replacing the hyperbolic tangent function in the additive attention mechanism with an activation function, and replacing the input of the additive attention mechanism with two feature layers of the same size but different semantic information, to obtain an improved additive attention mechanism; replacing the fusion method of the neck feature fusion layer in the YOLO model with the improved additive attention mechanism, and inputting the training dataset into the improved YOLO model for training to obtain the optimal weight model; inputting the aircraft skin damage image data to be detected into the model for detection, and obtaining the aircraft skin detection result. By combining the additive attention mechanism in the field of natural language processing with the neck feature fusion part of YOLO, an additive feature fusion attention mechanism is proposed, which makes the model pay more attention to the target part to be detected, while compressing the neck feature layer channels of the model, which can meet the requirements of high accuracy and lightweight, so as to better detect damage and fill the gap in related technologies. Therefore, it can effectively solve the problem of redundant model parameters in processing aircraft skin damage detection.

[0032] In another embodiment of the present invention, before conducting a simulation experiment on the optimal weight model to obtain the simulation test results, the method further includes: The feature layer is then subjected to visual analysis.

[0033] In this specific implementation, the feature layer visualization is GradCam feature layer visualization. This visualization analysis is mainly used to intuitively verify the role of the additive feature fusion attention mechanism.

[0034] In another embodiment of the present invention, the visualization analysis of the feature layer includes: The visualized sample data is input into the YOLO model and the optimal weight model respectively for forward propagation calculation, and the output results of the feature layer are recorded respectively; wherein the output results include activation maps; The activation maps generated by the YOLO model and the optimal weight model are visualized respectively. The analysis results are generated by comparing and analyzing activation maps, calculating difference images, or drawing histograms.

[0035] In this specific implementation, firstly, sample data for testing needs to be prepared. This data should include images of objects, which can be real-world images or synthetic images. Next, the YOLO model and the optimal weight model need to be loaded and compiled into executable models. Then, each sample data point is input into the YOLO model and the optimal weight model for forward propagation calculation. This will produce a series of feature layer outputs, including activation maps. If necessary, Keras prediction functions can be used to implement this process. The activation maps and feature layer outputs generated by each model are recorded. During recording, it is important to retain the label information of each sample data point for subsequent comparative analysis. Then, a Python visualization library such as Matplotlib is used to visualize the activation maps generated by each model. This allows for a better understanding of the model's detection results for different objects. Finally, the outputs of the two models are compared and analyzed by overlaying activation maps, calculating difference images, or drawing histograms. This allows for understanding the differences between the two models, identifying their strengths and weaknesses in the detection process, and guiding subsequent model improvements.

[0036] In another embodiment of the present invention, the improved additive attention mechanism includes at least two feature layers of the same size but different semantic information, a sigmoid activation function, and a Mut operation module; The feature layer is used to extract different semantic information from the input image; The sigmoid activation function is used to map the attention result to the [0,1] interval; The Mut operation module is used to calculate the similarity between the feature layers and to calculate the interaction weights between the feature layers based on the similarity.

[0037] In the specific implementation of this embodiment, please refer to Figure 3 This is a structural diagram of the improved additive feature fusion attention mechanism provided in this embodiment of the invention. The improvement aims to enhance the YOLO model's neck feature pyramid by replacing the original concat module with an additive feature fusion attention module. Specifically, the original concat feature fusion method is replaced with an add pixel-wise addition method, and an additive attention mechanism is fused. The hyperbolic tangent function in the additive attention mechanism is replaced with a sigmoid activation function, and the input is replaced with two feature layers of the same size but different semantic information. The Mut operation specifically involves pixel-wise multiplication of the feature layers. Through these improvements, the additive attention mechanism can better utilize the different information in the input features, improving the model's expressive power and generalization ability. Simultaneously, the introduction of the Mut operation module enhances the interaction between feature layers, helping to better capture the correlation between features.

[0038] In another embodiment provided by the present invention, the formula for the additive attention mechanism is:

[0039] In the formula, q represents the query vector, k represents the feature vector, and W v Let W represent a weight vector. q W represents the linear transformation matrix of the query vector. k Let represent the linear transformation matrix of the eigenvectors, tanh represent the hyperbolic tangent function, and T represent the linear transformation matrix; The formula for the improved additive attention mechanism is:

[0040] In the formula, `high_layer` represents high-level semantic information, `low_layer` represents low-level semantic information, and `sigmoid` represents the sigmoid activation function. Additive attention can be used to fuse multiple feature vectors into a single feature vector, focusing on those feature vectors that are most relevant to the query vector.

[0041] In another embodiment of the present invention, the step of evaluating the optimal weight model to obtain the evaluation test result includes: By inputting a preset test dataset into the optimal weight model, the model is tested and the model test results are generated. The model test results are compared and analyzed with the true label values ​​of the test data to obtain the model test error value; When the model test error value is less than the preset error value standard, the evaluation test result of the optimal weight model meets the preset project application requirements.

[0042] In this specific implementation, by inputting pre-set test data with known data components into the optimal weight model for testing, corresponding test results are generated, including the proportion of skin damage data in the test data, mAP (mean Average Precision), AP (Average Precision), precision, recall, test time, etc. Then, it is checked whether these indicators match the training results of the training set. If the difference meets the standard, it indicates that the model can be used and there is no overfitting. Next, it is checked whether these indicators of the model meet the practical application requirements. If they meet the standard, it can be determined that the optimal model meets the requirements, and the test results meet the pre-set project application needs. Model testing can further verify the model's ability to process data in practice, and by setting standards, models that meet the requirements can be selected.

[0043] In another embodiment of the present invention, the preprocessing of the aircraft skin damage image data includes: The aircraft skin damage image data was filtered to obtain aircraft skin damage image data in which the background of the images was entirely the aircraft fuselage. Then, remove aircraft skin damage image data from those where the number of targets exceeds the preset target number standard; Then, the aircraft skin damage image data is annotated using a preset data annotation tool.

[0044] In the specific implementation of this invention, by filtering and retaining images containing only the aircraft fuselage, the influence of other interfering objects on aircraft skin damage detection can be reduced. This improves the training and detection performance of subsequent models. By removing images with more than a preset standard of targets, it can be ensured that each image in the dataset contains only one aircraft skin damage target. This helps simplify the problem, making the model easier to learn and infer. By using a preset data annotation tool to annotate the aircraft skin damage image data with feature values, accurate label information can be provided for training the model. This helps the model understand and learn the characteristics of different types of aircraft skin damage and perform accurate detection and classification in practical applications. Through the above preprocessing steps, the dataset for the aircraft skin damage detection task can be optimized, improving the training effect and performance of the model. Operations such as filtering image backgrounds, removing multi-target images, and annotating feature values ​​help reduce noise and interference, improving the robustness and reliability of the model.

[0045] This invention also provides an aircraft skin damage detection system, see [link to relevant documentation]. Figure 2 This is a structural block diagram of an aircraft skin damage detection system provided in an embodiment of the present invention. The system includes: The data processing module 20 is used to acquire aircraft skin damage image data and preprocess the aircraft skin damage image data. The format conversion module 21 is used to convert the preprocessed aircraft skin damage image data into YOLO format data, and use the YOLO format data as a training dataset. The mechanism optimization module 22 is used to replace the hyperbolic tangent function in the additive attention mechanism with the sigmoid activation function, and to replace the input of the additive attention mechanism with two feature layers of the same size but different semantic information, so as to obtain the improved additive attention mechanism. Model optimization module 23 is used to obtain an improved YOLO model by replacing the concat feature fusion method of the neck feature fusion layer in the YOLO model with an improved additive attention mechanism; Model training module 24 is used to input the training dataset into the improved YOLO model for training to obtain the optimal weight model; Model testing module 25 is used to evaluate the optimal weight model and obtain evaluation test results; The aircraft skin damage detection module 26 is used to input the aircraft skin damage image data to the optimal weight model to perform aircraft skin damage detection if the evaluation test results meet the preset project application requirements, and obtain the aircraft skin damage detection results.

[0046] The system and modules described above can implement all the processes described in any of the above embodiments. The functions and technical effects of each module and unit in the system are the same as those of the aircraft skin damage detection method described in the above embodiments, and will not be repeated here.

[0047] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0048] In summary, this invention discloses an aircraft skin damage detection method, system, and storage medium, comprising: acquiring aircraft skin damage image data and preprocessing it; converting the preprocessed data to generate YOLO format data as a training dataset; replacing the hyperbolic tangent function in the additive attention mechanism with an activation function, and replacing the input of the additive attention mechanism with two feature layers of the same size but different semantic information to obtain an improved additive attention mechanism; replacing the fusion method of the neck feature fusion layer in the YOLO model with the improved additive attention mechanism, and inputting the training dataset into the improved YOLO model for training to obtain an optimal weight model; inputting the aircraft skin damage image data to be tested into the model for detection to obtain the aircraft skin detection result; effectively solving the problem of redundant model parameters in processing aircraft skin damage detection.

[0049] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for detecting damage to aircraft skin, characterized in that, The method includes: Acquire aircraft skin damage image data and preprocess the aircraft skin damage image data; The preprocessed aircraft skin damage image data is converted to YOLO format data, and the YOLO format data is used as the training dataset. The improved additive attention mechanism is obtained by replacing the hyperbolic tangent function in the additive attention mechanism with the sigmoid activation function and replacing the input of the additive attention mechanism with two feature layers of the same size but different semantic information. By replacing the concat feature fusion method in the neck feature fusion layer of the YOLO model with an improved additive attention mechanism, an improved YOLO model is obtained. The training dataset is input into the improved YOLO model for training to obtain the optimal weight model; The optimal weight model is evaluated to obtain the evaluation test results; If the evaluation test results meet the preset project application requirements, the aircraft skin damage image data to be tested is input into the optimal weight model to detect aircraft skin damage and obtain the aircraft skin detection results.

2. The method for detecting damage to aircraft skin as described in claim 1, characterized in that, Before conducting simulation experiments on the optimal weight model to obtain simulation test results, the following steps are also included: The feature layer is then subjected to visual analysis.

3. The method for detecting damage to aircraft skin as described in claim 2, characterized in that, The visualization analysis of the feature layer includes: The visualized sample data is input into the YOLO model and the optimal weight model respectively for forward propagation calculation, and the output results of the feature layer are recorded respectively; wherein the output results include activation maps; The activation maps generated by the YOLO model and the optimal weight model are visualized respectively. The analysis results are generated by comparing and analyzing activation maps, calculating difference images, or drawing histograms.

4. The method for detecting damage to aircraft skin as described in claim 1, characterized in that, The improved additive attention mechanism includes at least two feature layers of the same size but different semantic information, a sigmoid activation function, and a Mut operation module; The feature layer is used to extract different semantic information from the input image; The sigmoid activation function is used to map the attention result to the [0,1] interval; The Mut operation module is used to calculate the similarity between the feature layers and to calculate the interaction weights between the feature layers based on the similarity.

5. The method for detecting damage to aircraft skin as described in claim 1, characterized in that, The formula for the additive attention mechanism is: In the formula, q represents the query vector, k represents the feature vector, and W v Let W represent a weight vector. q W represents the linear transformation matrix of the query vector. k Let represent the linear transformation matrix of the eigenvectors, tanh represent the hyperbolic tangent function, and T represent the linear transformation matrix; The formula for the improved additive attention mechanism is: In the formula, high_layer represents high-level semantic information, low_layer represents low-level semantic information, and sigmoid represents the sigmoid activation function.

6. The method for detecting damage to aircraft skin as described in claim 1, characterized in that, The process of evaluating the optimal weight model to obtain evaluation test results includes: By inputting a preset test dataset into the optimal weight model, the model is tested and the model test results are generated. The model test results are compared and analyzed with the true label values ​​of the test data to obtain the model test error value; When the model test error value is less than the preset error value standard, the evaluation test result of the optimal weight model meets the preset project application requirements.

7. The method for detecting damage to aircraft skin as described in claim 1, characterized in that, The preprocessing of the aircraft skin damage image data includes: The aircraft skin damage image data was filtered to obtain aircraft skin damage image data in which the background of the images was entirely the aircraft fuselage. Then, remove aircraft skin damage image data from those where the number of targets exceeds the preset target number standard; Then, the aircraft skin damage image data is annotated using a preset data annotation tool.

8. An aircraft skin damage detection system, characterized in that, The system includes: The data processing module is used to acquire aircraft skin damage image data and preprocess the aircraft skin damage image data. The format conversion module is used to convert the preprocessed aircraft skin damage image data into YOLO format data, and use the YOLO format data as a training dataset. The mechanism optimization module is used to replace the hyperbolic tangent function in the additive attention mechanism with the sigmoid activation function, and to replace the input of the additive attention mechanism with two feature layers of the same size but different semantic information, so as to obtain the improved additive attention mechanism. The model optimization module is used to obtain an improved YOLO model by replacing the concat feature fusion method in the neck feature fusion layer of the YOLO model with an improved additive attention mechanism. The model training module is used to input the training dataset into the improved YOLO model for training to obtain the optimal weight model. The model testing module is used to evaluate the optimal weight model and obtain the evaluation test results. The aircraft skin damage detection module is used to input the image data of the aircraft skin damage to be tested into the optimal weight model to detect the aircraft skin damage if the evaluation test results meet the preset project application requirements, and obtain the aircraft skin damage detection results.

9. The aircraft skin damage detection system as described in claim 8, characterized in that, The system also includes a visualization module; The visualization module is used to perform visual analysis on the feature layer.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program; wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform an aircraft skin damage detection method as described in any one of claims 1 to 7.

Images