Honeycomb sandwich structure infrared image defect detection method and system based on improved YOLOv5

Abstract

The invention discloses a honeycomb sandwich structure infrared image defect detection method and system based on improved YOLOv5. The method comprises the following steps: collecting defect images of a honeycomb sandwich structure to form a sample data set; training defect images in the data set based on an improved YOLOv5 model, predicting positions and categories of honeycomb sandwich structure defects, training an optimization model by using a loss function until convergence, and obtaining and storing optimal model weight data; the improved YOLOv5 model is based on a YOLOv5 architecture, a Swin Transform module is embedded in a backbone network, and the detection performance of the model on honeycomb sandwich structure defects is improved in combination with a C3 convolutional layer, a CBS module and an SPPF module; and acquiring a defect image of the honeycomb sandwich structure to be detected, detecting and identifying the defect image by using the improved YOLOv5 model based on the optimal model weight data, and generating and outputting a detection result. According to the method, small-size defect identification in multi-scale defect identification can be effectively realized, and the defects of a traditional convolutional neural network in processing complex backgrounds and small target detection are overcome.

Description

Technical Field

[0001] This invention belongs to the field of honeycomb sandwich structure defect detection technology, specifically relating to a honeycomb sandwich structure infrared image defect detection method and system based on improved YOLOv5. Background Technology

[0002] The statements herein provide only background information in relation to this invention and do not necessarily constitute prior art.

[0003] In recent years, with the continuous advancement of national manned lunar landing, Mars exploration, and near-Earth spacecraft projects, higher requirements have been placed on the specific strength and specific modulus of materials used in launch vehicles and aerospace vehicles. Honeycomb sandwich composite materials, due to their high specific stiffness, high specific strength, low density, and excellent thermal insulation, sound insulation, and vibration damping properties, are widely used in primary and secondary load-bearing structural components of aerospace vehicles, such as aircraft wings, fuselage panels, radomes, and spines; spacecraft canopies, fairings, sections, and heat shields; and rocket and missile stabilizers, nose cones, and engine exhaust nozzles. However, due to the complex manufacturing process of honeycomb sandwich structures, and during the manufacturing process or long-term operation in harsh environments, defects such as debonding, collapse, skin delamination, water accumulation, and impact damage may occur. Among these, the most common defects are debonding between the skin and the adhesive layer, and between the adhesive layer and the honeycomb core. These defects or damages are typically small in area, discontinuous, and relatively hidden, meaning that honeycomb sandwich structures often fail without warning. However, they can suddenly collapse under external impact or internal stress, posing a fatal threat to the structure and severely impacting the normal use of related components. Therefore, for honeycomb sandwich structures, how to quickly, efficiently, and accurately detect and identify internal defects has become a research hotspot and challenge in the field of infrared nondestructive testing. The emergence of deep learning neural networks has provided a better solution for defect detection in honeycomb sandwich structures.

[0004] Currently, the detection of internal defects in cellular sandwich structures mainly employs single-stage and two-stage detection algorithms. Two-stage detection algorithms offer higher accuracy, but their complex model structure and high computational cost result in slow detection speeds, making them unsuitable for real-time detection. While single-stage detection algorithms offer speed advantages, their limited feature extraction capabilities lead to a significant drop in accuracy when detecting defects in complex backgrounds and small targets. The fundamental reason for this lies in the shortcomings of traditional convolutional neural networks in feature extraction and multi-scale information fusion, making it difficult to balance detection accuracy and real-time performance. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for detecting defects in infrared images of cellular sandwich structures based on an improved YOLOv5. The improved YOLOv5 detection algorithm, by introducing a Swing Transformer module into the backbone network, can effectively realize the identification of small-sized defects in multi-scale defect identification, and solves the shortcomings of traditional convolutional neural networks (CNNs) in handling complex backgrounds and small target detection.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution: On one hand, the technical solution of the present invention provides a method for detecting defects in infrared images of honeycomb sandwich structures based on improved YOLOv5, including: Defect images of honeycomb sandwich structures were collected to form a sample dataset; The improved YOLOv5 model is trained on defect images in the dataset to predict the location and category of defects in honeycomb sandwich structures. The model is trained and optimized until convergence using a loss function, and the optimal model weight data is obtained and saved. The improved YOLOv5 model is based on the YOLOv5 architecture, embedding a Swing Transformer module in the backbone network, and combining a C3 convolutional layer, a CBS module, and an SPPF module to improve the model's detection performance for defects in honeycomb sandwich structures. The defect image of the honeycomb sandwich structure to be detected is acquired. Based on the optimal model weight data, the improved YOLOv5 model is used to detect and identify the defect image of the honeycomb sandwich structure to be detected, and the detection results are generated and output.

[0007] In at least one embodiment, the process of collecting defect images of the honeycomb sandwich structure to form a sample dataset specifically involves: actively thermally stimulating the honeycomb sandwich structure sample with a laser, collecting defect image data of the honeycomb sandwich structure using an infrared camera, and classifying the defects of the honeycomb sandwich structure using manual annotation to form a sample dataset.

[0008] In at least one embodiment, the defect types of the honeycomb sandwich structure include debonding, cracks, voids, and delamination.

[0009] In at least one embodiment, Mosaic data augmentation is performed on the data during training to expand the amount of the dataset and reduce overfitting.

[0010] In at least one embodiment, the CBS module is composed of a convolutional layer, a batch normalization layer, and a ReLU activation function connected in sequence.

[0011] In at least one embodiment, the specific process of training the defective images in the dataset based on the improved YOLOv5 model includes: Defect images of the honeycomb sandwich structure are input into the improved YOLOv5 model. The defect images include different types of defects and complex backgrounds. The input defect image undergoes preliminary convolutional feature extraction and standardization through multiple CBD modules; Next, we use the C3+Swin module. The C3 module performs convolution for feature extraction, and the SwinTransformer module effectively captures long-range dependency information through self-attention operations of local windows. Then, the feature maps at different levels are merged using the Concat and Upsample modules; The SPPF module is introduced at the bottom of the model network to capture information at different scales in the defect image and perform multi-scale feature fusion. The final part of the model network includes multiple Conv2d and C3 modules to further extract and fuse features; Finally, the feature maps processed by the above modules are used by module C3 for defect prediction, generating the final detection results and outputting them through the output layer.

[0012] In at least one embodiment, the result of defect prediction may include the bounding box coordinates, confidence level, and class probability of defects in the cellular sandwich structure.

[0013] In at least one embodiment, the results obtained from defect prediction are processed using a nonmaximum suppression method to remove useless detection boxes, generate the final detection results, and output them.

[0014] In at least one embodiment, the output includes the category, bounding box coordinates, and confidence level for each detected target.

[0015] On the other hand, the technical solution of the present invention also provides an infrared image defect detection system for honeycomb sandwich structures based on the improved YOLOv5, including: The sample dataset generation module is configured to: acquire defect images of honeycomb sandwich structures to form a sample dataset; The model training module is configured to: train the defect images in the dataset based on the improved YOLOv5 model, predict the location and category of defects in the honeycomb sandwich structure, train and optimize the model until convergence using the loss function, and obtain and save the optimal model weight data; wherein, the improved YOLOv5 model is based on the YOLOv5 architecture, embeds the Swing Transformer module in the backbone network, and combines the C3 convolutional layer, CBS module and SPPF module to improve the model's detection performance of honeycomb sandwich structure defects; The defect detection module is configured to: acquire defect images of the honeycomb sandwich structure to be detected; based on the optimal model weight data, use the improved YOLOv5 model to detect and identify the defect images of the honeycomb sandwich structure to be detected; generate detection results and output them.

[0016] The beneficial effects of the above-described technical solution of the present invention are as follows: 1) This invention discloses an infrared image defect detection method for cellular sandwich structures based on an improved YOLOv5 algorithm. By introducing a Swing Transformer module into the backbone network, it effectively achieves small-sized defect identification in multi-scale defect recognition, overcoming the shortcomings of traditional convolutional neural networks (CNNs) in handling complex backgrounds and small target detection. This significantly improves the accuracy of small defect detection and reduces the false detection rate while maintaining high real-time performance. The self-attention mechanism of Swing Transformer effectively enhances the network's ability to identify complex defects. Furthermore, the use of the SPPF module and feature concatenation technology enables the model to process defects of different scales simultaneously, improving the detection capability for small target defects.

[0017] 2) The present invention provides an infrared image defect detection method for honeycomb sandwich structures based on improved YOLOv5, which is applicable to defect detection of honeycomb sandwich structures and is widely used in aerospace, shipbuilding, construction, energy and other fields. Its advantage is that it can efficiently and accurately detect various types of defects in complex environments and ensure structural safety. Attached Figure Description

[0018] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0019] Figure 1 This is a schematic diagram of the infrared image defect detection method for honeycomb sandwich structures based on the improved YOLOv5 disclosed in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the improved YOLOv5 model structure disclosed in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram comparing the results of multi-scale defect identification using the traditional YOLOv5 model and the improved YOLOv5 model of the present invention, as disclosed in Embodiment 1 of the present invention. (a) is a schematic diagram of a multi-size defect specimen, (b) is a schematic diagram of the defect identification results of the traditional YOLOv5 model, and (c) is a schematic diagram of the defect identification results of the improved YOLOv5 model of the present invention. Detailed Implementation

[0020] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0021] In existing research, Yang et al. modified the YOLOv3 feature pyramid network from unidirectional feature fusion to bidirectional fusion to improve the detection accuracy of small objects. Hao et al. introduced a cross-stage partial and residual segmentation attention network into the YOLOv4 backbone network, demonstrating enhanced feature extraction capabilities. They also adopted a bidirectional feature pyramid network with a simple attention mechanism to improve the accuracy of insulator defect detection in aerial images with complex background interference. However, although they improved the first-level detection algorithm to address the detection of small objects and challenging situations, this led to a significant increase in algorithm memory consumption and computational complexity. Xu et al. replaced the YOLOv4 backbone network with the lighter Mobilenet-V1 architecture and introduced a spatial and channel squeezing with channel activation (scSE) attention mechanism module to enhance the model's feature extraction capabilities, while combining depthwise separable convolutions to reduce the total number of network parameters. This method improved feature extraction capabilities while reducing network complexity, but excessive lightweighting led to a decrease in detection accuracy, especially in tasks involving complex backgrounds and small objects, where the recognition rate was insufficient, and the scSE module increased computational overhead to some extent. Guo et al. proposed MSFT-YOLO based on YOLOv5 for detecting small target defects on steel surfaces. They introduced a transformer-based Trans module into the backbone and neck network to fuse global information from feature maps, and combined it with a weighted bidirectional feature pyramid network to achieve effective fusion of information at different scales. This method improves the detection capability of small target defects, but the network structure is complex, the computational overhead for training and inference is high, and the hardware resource requirements are high, making it unsuitable for application in embedded devices or real-time detection scenarios. It also has a strong dependence on the number of samples and insufficient generalization performance. To improve the accuracy and speed of insulator defect detection, Han et al. designed the D-CSPDarknet53 network to enhance the YOLOv4 backbone network, reducing the model's parameters and computational complexity. They also incorporated SA-Net (Stochastic Attention Neural Network) into the feature fusion network to enhance target feature attention, and introduced multi-head output to improve the detection accuracy of small targets. This method improves detection accuracy, but with the increase in network structure, the overall model complexity and training cost increase significantly, the inference speed decreases, and its adaptability to complex noise environments and small sample conditions is insufficient.

[0022] To overcome the shortcomings of the existing technology, the present invention provides a method and system for detecting defects in infrared images of cellular sandwich structures based on improved YOLOv5. Based on the improved YOLOv5 detection algorithm, by introducing the Swing Transformer module into the backbone network, it can effectively realize the identification of small-sized defects in multi-scale defect identification, and solve the shortcomings of traditional convolutional neural networks (CNN) in handling complex backgrounds and small target detection.

[0023] Example 1 In a typical embodiment of the present invention, such as Figures 1 to 3 As shown, this embodiment discloses a method for detecting defects in cellular sandwich structures using infrared images based on an improved YOLOv5, including the following steps: S100. Collect defect images of the honeycomb sandwich structure to form a sample dataset; S200. Based on the improved YOLOv5 model, the defect images in the dataset are trained to predict the location and category of defects in the honeycomb sandwich structure. The loss function is used to train and optimize the model until convergence, and the optimal model weight data is obtained and saved. Among them, the improved YOLOv5 model is based on the YOLOv5 architecture, embedding the Swing Transformer module in the backbone network, and combining the C3 convolutional layer, CBS module and SPPF module to improve the model's detection performance of honeycomb sandwich structure defects. S300. Acquire the defect image of the honeycomb sandwich structure to be detected, and based on the optimal model weight data, use the improved YOLOv5 model to detect and identify the defect image of the honeycomb sandwich structure to be detected, generate detection results and output them.

[0024] The following detailed description of the infrared image defect detection method for honeycomb sandwich structures based on the improved YOLOv5 is provided in conjunction with specific implementation methods.

[0025] S100. Collect defect images of the honeycomb sandwich structure to form a sample dataset.

[0026] In this embodiment, the honeycomb sandwich structure sample is first subjected to active thermal excitation using a laser, and defect image data of the honeycomb sandwich structure is acquired using an infrared camera or infrared thermal imager. The defect image data contains different types of defects, including debonding, cracks, voids, and delamination. Then, the location, type, and severity of the defects in the honeycomb sandwich structure are labeled manually, thereby classifying the defects in the honeycomb sandwich structure and forming a sample dataset for subsequent model training and testing.

[0027] In this embodiment, after forming the sample dataset, the data in the sample dataset is divided into a training set, a validation set, and a test set in a ratio of 7:2:1, which are used for training and testing the model, respectively.

[0028] S200. Based on the improved YOLOv5 model, the defect images in the dataset are trained to predict the location and category of defects in the honeycomb sandwich structure. The model is trained and optimized until convergence using a loss function, and the final model weight data is obtained and saved. The improved YOLOv5 model is based on the YOLOv5 architecture, embedding the Swing Transformer module in the backbone network, and combining the C3 convolutional layer, CBS module and SPPF module to improve the model's detection performance of defects in the honeycomb sandwich structure.

[0029] In this embodiment, the training set from the sample dataset is input into the improved YOLOv5 model for training. During training, the Mosaic data augmentation method is used to augment the data, expanding the dataset and reducing overfitting. Furthermore, by randomly using four images, randomly scaling them, and then randomly distributing and stitching them together, the detection dataset is greatly enriched. In particular, the random scaling adds many small targets, which improves the robustness of the model network.

[0030] like Figure 2 As shown, the network structure proposed in this embodiment is based on the YOLOv5 architecture. A Swing Transformer module is embedded in the backbone network, and the C3 convolutional layer, CBS module, and SPPF module are combined to improve the model's detection performance for defects in cellular sandwich structures. By combining multiple components such as the Swing Transformer module, convolutional layer (C3), and CBS module, the improved model is ensured to achieve efficient feature extraction and multi-scale processing.

[0031] The specific training process is as follows: (1) Input images: Defect images of honeycomb sandwich structures in the training set are used as input to the model network. These images may include different types of defects (such as cracks, voids, etc.) and contain complex backgrounds.

[0032] (2) CBS Module: The input image first passes through multiple CBS modules for preliminary convolutional feature extraction and standardization. This process can effectively extract basic image features. The CBS module consists of convolutional layers, batch normalization layers, and ReLU activation functions connected in sequence.

[0033] (3) C3+Swin Module: The subsequent network layers of the model use the C3+Swin module, where C3 is a convolutional layer used for feature extraction, and the Swin Transformer module introduces a self-attention mechanism into the model during this process. The Swin Transformer can effectively capture long-range dependency information through the self-attention operation of local windows, enhancing its ability to handle complex defects and small targets. Through this structure, the network can process features at multiple levels, maintain the details of the information, and capture global features at the same time, which is particularly suitable for small cracks or hard-to-detect defects in honeycomb sandwich structures.

[0034] (4) Concat and Upsample: Feature concatenation is used to merge feature maps at different levels to further enhance the model's ability to detect defects at different scales. Upsampling helps the model restore the resolution of the feature maps for more precise localization and detection.

[0035] (5) SPPF module: At the bottom of the model network, the SPPF (Spatial Pyramid Pooling) module is introduced. It is used for multi-scale feature fusion and can effectively capture information of different scales in the image, so that the model can better handle multi-scale defects.

[0036] (6) Conv2d and C3 modules: The last part of the network includes multiple convolutional operations (Conv2d) and C3 modules, which further extract and fuse features to ensure the accuracy and precision of the detection results.

[0037] (7) Output layer: Finally, the feature maps processed by the above modules are passed through module C3 to generate the final detection results. The output includes the category, bounding box coordinates, and confidence score of each detected target.

[0038] (8) Train and optimize the model using the loss function until convergence, and evaluate the model's performance using the validation set during training. Save the model weight parameter file with the best accuracy on the validation set. The loss function consists of bounding box regression loss, confidence loss, and classification loss, which are used to constrain the positional relationship between the predicted box and the ground truth box, the probability of the target's existence, and the identification of the defect category. During training, the model continuously updates the network parameters through forward propagation and backpropagation combined with gradient descent until it achieves optimal performance on the validation set. The optimal model weight data is then obtained and saved.

[0039] S300. Acquire the defect image of the honeycomb sandwich structure to be detected, and based on the optimal model weight data, use the improved YOLOv5 model to detect and identify the defect image of the honeycomb sandwich structure to be detected, generate detection results and output them.

[0040] The optimal model weight data obtained from training in step S200 is loaded. The defect image of the honeycomb sandwich structure to be detected (i.e., the test set data) is input into the model. After feature extraction and aggregation, the bounding box coordinates, confidence scores, and class probabilities that may contain honeycomb sandwich structure defects are obtained. The results of defect prediction are processed using the non-maximum suppression method to remove useless detection boxes, and the final detection results are generated and output. The output includes the class, bounding box coordinates, and confidence score of each detected target.

[0041] This embodiment presents a method for detecting defects in infrared images of cellular sandwich structures based on an improved YOLOv5 algorithm. By introducing a Swing Transformer module into the backbone network, it addresses the shortcomings of traditional convolutional neural networks (CNNs) in handling complex backgrounds and small target detection. Figure 3 As shown, for Figure 3 (a) The specimen with dimensional defects shown in the figure. Figure 3 (b) The traditional YOLOv5 cannot identify small-sized defects when performing multi-scale defect identification, while Figure 3 (c) The improved YOLOv5 proposed in this embodiment can effectively identify small-sized defects when performing multi-scale defect recognition. The self-attention mechanism of the Swin Transformer can effectively improve the network's ability to identify complex defects. Moreover, the use of SPPF modules and feature stitching technology enables the model to process defects of different scales simultaneously, improving the detection capability of small target defects. This method is suitable for defect detection in cellular sandwich structures and is widely used in aerospace, shipbuilding, construction, energy and other fields. Its advantage lies in its ability to efficiently and accurately detect various types of defects in complex environments, ensuring structural safety.

[0042] Example 2 In a typical embodiment of the present invention, this embodiment discloses an infrared image defect detection system for honeycomb sandwich structures based on an improved YOLOv5, comprising: The sample dataset generation module is configured to: acquire defect images of honeycomb sandwich structures to form a sample dataset; The model training module is configured to: train the defect images in the dataset based on the improved YOLOv5 model, predict the location and category of defects in the honeycomb sandwich structure, train and optimize the model until convergence using the loss function, and obtain and save the optimal model weight data; wherein, the improved YOLOv5 model is based on the YOLOv5 architecture, embeds the Swing Transformer module in the backbone network, and combines the C3 convolutional layer, CBS module and SPPF module to improve the model's detection performance of honeycomb sandwich structure defects; The defect detection module is configured to: acquire defect images of the honeycomb sandwich structure to be detected; based on the optimal model weight data, use the improved YOLOv5 model to detect and identify the defect images of the honeycomb sandwich structure to be detected; generate detection results and output them.

[0043] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting defects in honeycomb sandwich structures using infrared images based on an improved YOLOv5, characterized in that, include: Defect images of honeycomb sandwich structures were collected to form a sample dataset; The improved YOLOv5 model is trained on defect images in the dataset to predict the location and category of defects in honeycomb sandwich structures. The model is trained and optimized until convergence using a loss function, and the optimal model weight data is obtained and saved. The improved YOLOv5 model is based on the YOLOv5 architecture, embedding a Swing Transformer module in the backbone network, and combining a C3 convolutional layer, a CBS module, and an SPPF module to improve the model's detection performance for defects in honeycomb sandwich structures. The defect image of the honeycomb sandwich structure to be detected is acquired. Based on the optimal model weight data, the improved YOLOv5 model is used to detect and identify the defect image of the honeycomb sandwich structure to be detected, and the detection results are generated and output.

2. The method for detecting defects in honeycomb sandwich structures using infrared images based on improved YOLOv5 as described in claim 1, characterized in that, The specific steps for collecting defect images of the honeycomb sandwich structure to form a sample dataset are as follows: actively thermally excite the honeycomb sandwich structure sample using a laser, collect defect image data of the honeycomb sandwich structure using an infrared camera, and classify the defects of the honeycomb sandwich structure using manual annotation to form a sample dataset.

3. The infrared image defect detection method for honeycomb sandwich structures based on improved YOLOv5 as described in claim 2, characterized in that, Defect types in honeycomb sandwich structures include debonding, cracks, voids, and delamination.

4. The infrared image defect detection method for honeycomb sandwich structures based on improved YOLOv5 as described in claim 1, characterized in that, Mosaic data augmentation is applied to the data during training to expand the dataset and reduce overfitting.

5. The infrared image defect detection method for honeycomb sandwich structures based on improved YOLOv5 as described in claim 1, characterized in that, The CBS module consists of a convolutional layer, a batch normalization layer, and a ReLU activation function connected in sequence.

6. The method for detecting defects in honeycomb sandwich structures using infrared images based on improved YOLOv5 as described in claim 1, characterized in that, The specific process of training the defective images in the dataset based on the improved YOLOv5 model includes: Defect images of the honeycomb sandwich structure are input into the improved YOLOv5 model. The defect images include different types of defects and complex backgrounds. The input defect image undergoes preliminary convolutional feature extraction and standardization through multiple CBD modules; Next, we use the C3+Swin module. The C3 module performs convolution for feature extraction, and the Swin Transformer module effectively captures long-range dependency information through self-attention operations of local windows. Then, the feature maps at different levels are merged using the Concat and Upsample modules; The SPPF module is introduced at the bottom of the model network to capture information at different scales in the defect image and perform multi-scale feature fusion. The final part of the model network includes multiple Conv2d and C3 modules to further extract and fuse features; Finally, the feature maps processed by the above modules are used by module C3 for defect prediction, generating the final detection results and outputting them through the output layer.

7. The infrared image defect detection method for honeycomb sandwich structures based on improved YOLOv5 as described in claim 6, characterized in that, The results of defect prediction may include the bounding box coordinates, confidence level, and class probability of defects in cellular sandwich structures.

8. The infrared image defect detection method for honeycomb sandwich structures based on improved YOLOv5 as described in claim 7, characterized in that, The results obtained from defect prediction are processed using the nonmaximum suppression method to remove useless detection boxes, generate the final detection results, and output them.

9. The infrared image defect detection method for honeycomb sandwich structures based on improved YOLOv5 as described in claim 8, characterized in that, The output includes the category, bounding box coordinates, and confidence score for each detected target.

10. A honeycomb sandwich structure infrared image defect detection system based on improved YOLOv5, characterized in that, include: The sample dataset generation module is configured to: acquire defect images of honeycomb sandwich structures to form a sample dataset; The model training module is configured to: train the defect images in the dataset based on the improved YOLOv5 model, predict the location and category of defects in the honeycomb sandwich structure, train and optimize the model until convergence using the loss function, and obtain and save the optimal model weight data; wherein, the improved YOLOv5 model is based on the YOLOv5 architecture, embeds the SwinTransformer module in the backbone network, and combines the C3 convolutional layer, CBS module and SPPF module to improve the model's detection performance of honeycomb sandwich structure defects; The defect detection module is configured to: acquire defect images of the honeycomb sandwich structure to be detected; based on the optimal model weight data, use the improved YOLOv5 model to detect and identify the defect images of the honeycomb sandwich structure to be detected; generate detection results and output them.

Images