Aero-engine skin defect detection method and system
The aero-engine skin defect detection method constructed by neural networks and mask scoring R-CNN algorithm solves the problems of long detection time and low efficiency in existing technologies, realizes high-precision skin defect detection, and improves the intelligence and informatization level of aero-engines.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHEJIANG GONGSHANG UNIVERSITY
- Filing Date
- 2023-03-20
- Publication Date
- 2026-06-23
AI Technical Summary
Existing technologies for detecting defects in aero-engine skin are time-consuming, inefficient, and lack specificity, affecting the overall maintenance efficiency and safety of aircraft.
A defect detection method for aero-engine skin based on neural networks and mask scoring R-CNN algorithm is adopted. The defect detection model is constructed through preprocessing, feature parameter extraction and mask scoring R-CNN training. A convolutional block attention module and a feature fusion module are introduced to improve the detection accuracy.
It has achieved pixel-level detection of defects in the skin of aero-engines, improved the accuracy and intelligence of defect identification, shortened maintenance time, and enhanced detection precision and generalization ability.
Smart Images
Figure CN116523834B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of defect detection technology based on aero-engine skin, and in particular to a method and system for detecting defects in aero-engine skin. Background Technology
[0002] Aircraft skin often suffers from surface cracks, impacts, and other defects, which not only affect the aircraft's appearance but also, to some extent, the health of internal components and the aircraft's lifespan. In some cases, they can even endanger the safety of passengers. Therefore, it is crucial to be able to identify surface defects in a timely manner and dynamically monitor the aircraft's health status.
[0003] Aircraft skin inspection includes detecting surface defects caused by corrosion, cracks, and stains, such as oil spills, grease, and dirt deposits. Conventional skin inspections are mostly done manually, which is time-consuming and inefficient. Currently, the aviation industry focuses on overall aircraft health monitoring for skin inspections. As a core component of civil aircraft, the aero-engine involves many parts and is highly susceptible to various types of damage. However, there are no specific methods for detecting these skin defects. If rapid identification and defect detection could be achieved for each aero-engine component, simplifying the defect detection process and improving accuracy, it would significantly improve overall aircraft maintenance efficiency, shorten crew overhaul time, and provide a milestone in aviation informatization and digitalization. Summary of the Invention
[0004] The purpose of this application is to overcome the shortcomings of the existing technology and provide a method and system for detecting defects in the skin of an aero-engine.
[0005] Firstly, a method for detecting defects in the skin of an aero-engine is provided, including:
[0006] Acquire images of the aircraft engine skin;
[0007] The image is preprocessed to remove image noise;
[0008] Constructing an aircraft engine skin defect detection model: Extracting feature parameters of the aircraft engine skin surface under various aircraft engine defects through a pre-trained neural network, constructing a set of aircraft engine skin defect feature parameters, and then training several defect classification sets through the mask scoring R-CNN algorithm to form an aircraft engine skin defect knowledge base;
[0009] The preprocessed image to be tested is input into the aero-engine skin defect detection model to output aero-engine defect status information.
[0010] Furthermore, the preprocessing employs median filtering to remove image noise.
[0011] Furthermore, a convolutional block attention module and a feature fusion module are added to the neural network. The convolutional block attention module includes a channel attention submodule and a spatial attention submodule. The channel attention submodule is used to mine the inter-channel relationships of features and filter out feature maps with strong discriminative capabilities. The spatial attention submodule focuses on the position of image information parts and constructs a rule set of feature parameters for aircraft engine skin defects.
[0012] Furthermore, the aircraft engine skin defect detection model includes:
[0013] The feature parameter extraction module is used to extract feature parameters of the aero-engine skin surface under various aero-engine defects through a pre-trained neural network, and to construct a rule set of aero-engine skin defect feature parameters;
[0014] The mask scoring R-CNN module is used for region candidate network processing, generating suggestions describing regions that may contain targets, and performing multi-class classification, candidate box regression, and introducing a fully convolutional neural network to generate masks for regions that may contain targets.
[0015] Output module, used to output aircraft engine defects.
[0016] Furthermore, the mask scoring R-CNN module includes:
[0017] A region candidate network is used to generate suggestions describing regions that may contain the target.
[0018] The classifier header is used to regress bounding box coordinates and class associations.
[0019] Masking tool, used for binary mask calculation;
[0020] The mask-iou header is used for mask scoring evaluation.
[0021] Furthermore, the classifier head includes four convolutional layers and one fully connected layer, and the loss function is:
[0022]
[0023] in, It is the loss of the RPN classification task, J cls The loss function used in masked R-CNN is the cross-entropy loss function. and J bbox These are the regression loss functions for RPN and Mask R-CNN, respectively, using a smoothed L1 loss function, J maskIoU It is the Mask-IoU regression loss, using the L2 loss function.
[0024] Secondly, a system for identifying defects in the skin of an aircraft engine is provided, including:
[0025] The image acquisition module is used to acquire images of the aircraft engine skin;
[0026] A preprocessing module is used to preprocess the image to remove image noise;
[0027] The model building module is used to build an aircraft engine skin defect detection model: it extracts the feature parameters of the aircraft engine skin surface under various aircraft engine defects through a pre-trained neural network, builds a set of aircraft engine skin defect feature parameters, and then trains several defect classification sets through the mask scoring R-CNN algorithm to form an aircraft engine skin defect knowledge base.
[0028] The output module outputs the defect status information of the aero-engine.
[0029] Furthermore, the aircraft engine skin defect detection model includes:
[0030] The feature parameter extraction module is used to extract feature parameters of the aero-engine skin surface under various aero-engine defects through a pre-trained neural network, and to construct a rule set of aero-engine skin defect feature parameters. The neural network includes a convolutional block attention module and a feature fusion module.
[0031] The mask scoring R-CNN module is used for region candidate network processing, generating suggestions describing regions that may contain targets, and performing multi-class classification, candidate box regression, and introducing a fully convolutional neural network to generate masks for regions that may contain targets.
[0032] Output module, used to output aircraft engine defects.
[0033] Furthermore, the mask scoring R-CNN module includes:
[0034] A region candidate network is used to generate suggestions describing regions that may contain the target.
[0035] The classifier header is used to regress bounding box coordinates and class associations.
[0036] Masking tool, used for binary mask calculation;
[0037] The mask-iou header is used for mask scoring evaluation.
[0038] Furthermore, the classifier head includes four convolutional layers and one fully connected layer, and the loss function is:
[0039]
[0040] in, It is the loss of the RPN classification task, J cls The loss function used in masked R-CNN is the cross-entropy loss function. and J bbox These are the regression loss functions for RPN and Mask R-CNN, respectively, using a smoothed L1 loss function, J maskIoU It is the Mask-IoU regression loss, using the L2 loss function.
[0041] Thirdly, a computer-readable storage medium is provided that stores program code for execution by a device, the program code including steps for performing a method as described in any implementation of the first aspect.
[0042] Fourthly, an electronic device is provided, the electronic device including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the method as in any of the implementations of the first aspect.
[0043] This application has the following beneficial effects: Compared with the prior art, the technical solution of this application can realize pixel-level detection of defects in the skin of aero-engines. Under the mask scoring R-CNN framework, the design of this technical solution introduces a new classifier head, attention mechanism and feature fusion module to improve the accuracy of defect detection. It integrates defect recognition experience with machine learning. Compared with traditional machine vision detection, it has higher precision and accuracy, significantly improves the level of intelligence and informatization, and has greater generalization ability and application prospects. Attached Figure Description
[0044] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an undue limitation of this application.
[0045] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0046] Figure 1 This is a flowchart of the aircraft engine skin defect detection method according to Embodiment 1 of this application;
[0047] Figure 2 This is a schematic diagram of the structure of the aircraft engine skin defect detection model in the aircraft engine skin defect detection method of Embodiment 1 of this application;
[0048] Figure 3 This is a flowchart of image preprocessing in the aircraft engine skin defect detection method of Embodiment 1 of this application;
[0049] Figure 4 This is a diagram showing the result of defect detection in the aircraft engine skin defect detection method of Embodiment 1 of this application;
[0050] Figure 5 This is a structural block diagram of the aircraft engine skin defect detection system according to Embodiment 1 of this application. Detailed Implementation
[0051] 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.
[0052] Example 1
[0053] This application discloses an aero-engine skin defect detection method, comprising: acquiring an image of the aero-engine skin; preprocessing the image to remove image noise; constructing an aero-engine skin defect detection model: extracting feature parameters of the aero-engine skin surface under various aero-engine defects through a pre-trained neural network, constructing a rule set of aero-engine skin defect feature parameters, and then training several defect classification sets through a mask scoring R-CNN algorithm to form an aero-engine skin defect knowledge base; inputting the preprocessed image to be tested into the aero-engine skin defect detection model to output aero-engine defect status information. This method can achieve pixel-level detection of aero-engine skin defects. Under the mask scoring R-CNN framework, the design of this technical solution introduces a new classifier head, attention mechanism, and feature fusion module to improve the accuracy of defect detection, integrates defect recognition experience with machine learning, and compared with traditional machine vision detection, has higher accuracy and guaranteed precision, significantly improved intelligence and informatization level, and has greater generalization ability and application prospects.
[0054] Specifically, Figure 1 A flowchart of the aircraft engine skin defect detection method in Embodiment 1 of the application is shown, including:
[0055] S101. Obtain an image of the aircraft engine skin;
[0056] For example, the image of the aircraft engine skin can be an image of the aircraft generator skin taken by electronic devices such as cameras, surveillance cameras, and mobile phones, and the collected real-time, dynamic engine skin damage image can be transmitted wirelessly to the corresponding engine health monitoring platform.
[0057] S102. Preprocess the image to remove image noise;
[0058] Specifically, such as Figure 3 As shown, preprocessing is performed on the skin image of the aircraft engine to make the defect detection more representative. Through experimental comparison, the median filtering method is finally selected for image preprocessing to remove image noise and improve the accuracy of the data. The median filtering method is a non-linear smoothing technique that sets the gray value of each pixel to the median of the gray values of all pixels in a certain neighborhood window of that point. Since the median filtering method is a conventional technique in the field of image processing technology, the specific processing flow will not be described in detail here.
[0059] S103. Construct an aircraft engine skin defect detection model: Extract feature parameters of the aircraft engine skin surface under various aircraft engine defects through a pre-trained neural network, construct an aircraft engine skin defect feature parameter rule set, and then train several defect classification sets through the mask scoring R-CNN algorithm to form an aircraft engine skin defect knowledge base.
[0060] Please see Figure 2 After processing images of aircraft engine skin defect detection, to better extract feature parameters and learn more important features for defect detection, a convolutional block attention module and a feature fusion module were added to the mask scoring R-CNN framework. The feature maps output from the backbone network were sequentially input into the two newly added modules (i.e., the convolutional block attention module and the feature fusion module), enhancing the ability to identify defects, ignore redundant and invalid information, and fuse features at different scales. Then, a pre-trained neural network was used to extract feature parameters of the aircraft engine skin surface under various aircraft engine defects, and an aircraft engine skin defect feature parameter specification was constructed. After feature extraction, the dataset is processed by a Region Proposal Network (RPN) to generate proposals describing regions that may contain targets, called Regions of Interest (ROIs). Following the ROI information, there are three branches. The first branch is a classifier head, which is used to regress bounding box coordinates and classify the associated class. It can also reduce the influence of surrounding information on the proposal, improving the accuracy of defect region localization and defect category recognition. This classifier head contains 4 convolutional layers and 1 fully connected layer. In the second branch, the object is segmented at the pixel level, and a mask head is used for binary mask calculation. The third branch is a mask-iou head, which is used for mask scoring evaluation.
[0061] Specifically, the convolutional block attention module includes a channel attention submodule and a spatial attention submodule. The channel attention submodule is used to mine the inter-channel relationships of features and filter out feature maps with strong discriminative capabilities. The spatial attention submodule focuses on the location of image information parts and constructs a set of feature parameter rules for aero-engine skin defects.
[0062] The channel attention submodule is used to mine the inter-channel relationships of features. It filters out feature maps with strong discriminative power. Specifically, it assumes that the feature map F∈R... m×n×c For the output of the backbone network, for each channel F i ∈R m×n The feature maps of i = 1, 2, ... are processed using global max pooling and global average pooling operations; two feature vectors are generated respectively: max pooling feature vector and average pooling feature vector. and average pool characteristics The two vectors MC and AC are then sent to a multilayer perceptron, which consists of a C-dimensional input layer, an R-dimensional hidden layer (where R = C / 4), and a C-dimensional output module. The two new feature vectors M'C and A'C output by the multilayer perceptron are combined using element-wise summation, and then normalized using the sigmoid function to obtain the channel attention weight vector W. c ∈R c Each element of the channel attention weight vector is multiplied by the corresponding channel's feature map to obtain a new feature representation F'∈R. m×n×c .
[0063] For the spatial attention submodule, which focuses on the location of image information, F' represents the input of this submodule. First, max pooling and average pooling operations are used to calculate the maximum and average values along the channel axis, respectively, to obtain two matrices. Second, these two matrices are concatenated to generate a feature tensor. Finally, the convolutional network uses the sigmoid function to generate the final spatial attention weight matrix.
[0064] Regarding the design of the classifier head, it plays a crucial role in the Mask Scoring R-CNN framework. It provides the category of each object and the location of its corresponding bounding box. The size of the bounding box is the most important factor in determining which layer's feature map is used to represent the object's features. It also affects the accuracy of defect region segmentation and the object category used to select the final object mask channel. The original classifier head of Mask Scoring R-CNN consists of two cascaded fully connected layers, followed by two parallel fully connected layers, one for object classification and the other for bounding box regression. Fully connected (FC) layers are more spatially sensitive than convolutional layers and are suitable for determining object categories, while convolutional layers are better suited for localization. In regression tasks, when fully connected (FC) layers are connected to convolutional layers, the information from the convolutional layers first needs to be flattened into vectors in a certain order, and then used as input to the fully connected (FC) layers. Therefore, the fully connected (FC) layers can combine the local information in the convolutional layers with category recognition, making them suitable for object classification. This method uses four concatenated convolutional layers to re-encode the feature maps corresponding to the bounding boxes. The output sizes of the convolutional layers are the same. Then, the re-encoded feature maps are mapped to a vector through the fully connected (FC) layers. Finally, two parallel fully connected layers are used: one for object classification and the other for bounding box regression, ultimately forming a more accurate model of aircraft engine skin defects.
[0065] For example, the aircraft engine skin defect detection model includes:
[0066] The feature parameter extraction module is used to extract feature parameters of the aero-engine skin surface under various aero-engine defects through a pre-trained neural network, and to construct a rule set of aero-engine skin defect feature parameters. The purpose of the feature parameter extraction module is to generate feature maps, and the feature maps are directly extracted by the pre-trained neural network.
[0067] The Mask Scoring R-CNN module is used for region candidate network processing to generate suggestions describing regions that may contain targets, and performs multi-class classification, candidate box regression, and introduces a fully convolutional neural network (FCN) to generate masks for regions that may contain targets (i.e., regions of interest).
[0068] Output module, used to output aircraft engine defects.
[0069] In a further embodiment, the mask scoring R-CNN module includes:
[0070] The Region Candidate Network (RPN) is used to generate suggestions describing regions that may contain the target, called regions of interest (ROIs).
[0071] The classifier header is used to regress bounding box coordinates and class associations.
[0072] Masking tool, used for binary mask calculation;
[0073] The mask-iou header is used for mask scoring evaluation.
[0074] It should be noted that the basic framework of this aircraft engine skin defect detection model is inherited from Mask R-CNN. The aircraft engine skin defect detection model is an improved version of Mask R-CNN, which adds a branch - called the Mask-iou head - to learn the mask score aligned with the Mask-iou. In addition, in order to enhance the expressive power of the feature map, a Convolutional Block Attention Module is added to the algorithm.
[0075] In a further embodiment, the classifier head includes four convolutional layers and one fully connected layer, and the loss function is:
[0076]
[0077] in, It is the loss of the RPN classification task, J cls The loss function used in masked R-CNN is the cross-entropy loss function. and J bbox These are the regression loss functions for RPN and Mask R-CNN, respectively, using a smoothed L1 loss function, J maskIoU It is the Mask-IoU regression loss, using the L2 loss function.
[0078] S104. Input the preprocessed image to be tested into the aero-engine skin defect detection model to output aero-engine defect status information.
[0079] For details, please refer to Figure 4 , Figure 4From left to right, the image displays the detection results for engine defects including surface foreign objects, engine scratches, skin stains, and skin stain defects. Based on the different input images of the aircraft engine skin, the above design outputs the engine defect status, achieving aircraft engine defect detection. This is mainly divided into four modules: the first module is the aircraft engine image acquisition and processing module; the second module is the feature parameter extraction module, whose purpose is to generate feature maps, which are directly extracted by a pre-trained neural network; the third module is the mask scoring R-CNN module, including a region candidate network (RPN), used to generate suggestions describing regions that may contain targets, called regions of interest (ROIs); and the fourth module is the output module, which outputs the defect type according to the corresponding defect classification data rules, achieving the ultimate goal of engine skin defect detection. This means detecting and identifying aircraft engine skin images through a trained model and accurately classifying skin defect types, mainly into five categories: skin cracks, component corrosion, holes caused by accidental impacts, paint peeling, and skin deformation.
[0080] Through quantitative comparison, the method proposed in this application... It achieved best performance across six metrics, indicating small detection and segmentation speeds. middle and large Compared with masked R-CNN, this method improves the APB and APM by approximately 27.66% and 20.49%, respectively. The improved method's ability to detect multi-scale targets makes a significant contribution to the successful detection of aircraft skin defects. Experimental results show that, compared with traditional methods, the improved algorithm and the addition of an attention mechanism improve the average detection accuracy by 16.98%, and significantly reduce the false negative and false positive rates, effectively verifying the practicality and accuracy of the proposed method.
[0081] Example 2
[0082] The second embodiment of this application relates to an identification system for detecting defects in the skin of an aircraft engine, such as... Figure 5 As shown, it includes:
[0083] The image acquisition module is used to acquire images of the aircraft engine skin;
[0084] A preprocessing module is used to preprocess the image to remove image noise;
[0085] The model building module is used to build an aircraft engine skin defect detection model: it extracts the feature parameters of the aircraft engine skin surface under various aircraft engine defects through a pre-trained neural network, builds a set of aircraft engine skin defect feature parameters, and then trains several defect classification sets through the mask scoring R-CNN algorithm to form an aircraft engine skin defect knowledge base.
[0086] The output module outputs the defect status information of the aero-engine.
[0087] The aircraft engine skin defect detection model includes:
[0088] The feature parameter extraction module is used to extract feature parameters of the aero-engine skin surface under various aero-engine defects through a pre-trained neural network, and to construct a rule set of aero-engine skin defect feature parameters. The neural network includes a convolutional block attention module and a feature fusion module.
[0089] The mask scoring R-CNN module is used for region candidate network processing, generating suggestions describing regions that may contain targets, and performing multi-class classification, candidate box regression, and introducing a fully convolutional neural network to generate masks for regions that may contain targets.
[0090] Output module, used to output aircraft engine defects.
[0091] The mask scoring R-CNN module includes:
[0092] A region candidate network is used to generate suggestions describing regions that may contain the target.
[0093] The classifier header is used to regress bounding box coordinates and class associations.
[0094] Masking tool, used for binary mask calculation;
[0095] The mask-iou header is used for mask scoring evaluation.
[0096] Example 3
[0097] The present application discloses a computer-readable storage medium that stores program code for execution by a device, the program code including steps for performing the method in any implementation of the present application, as described in the first embodiment of the present application.
[0098] The computer-readable storage medium may be a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM); the computer-readable storage medium may store program code, and when the program stored in the computer-readable storage medium is executed by a processor, the processor is used to perform the steps of the method in any of the implementations of Embodiment 1 of this application.
[0099] Example 4
[0100] An electronic device according to Embodiment 4 of this application includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, they implement the method in any of the implementations in Embodiment 1 of this application.
[0101] The processor can be a general-purpose central processing unit (CPU), microprocessor, application-specific integrated circuit (ASIC), graphics processing unit (GPU), or one or more integrated circuits, used to execute relevant programs to implement the method in any of the implementations of Embodiment 1 of this application.
[0102] The processor can also be an integrated circuit electronic device with signal processing capabilities. In implementation, each step of the method in any of the implementations of Embodiment 1 of this application can be completed by the integrated logic circuitry in the processor's hardware or by software instructions.
[0103] The aforementioned processor can also be a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the functions required by the units included in the data processing apparatus of the embodiments of this application, or executes the methods in any implementation of Embodiment 1 of this application.
[0104] The above are merely preferred embodiments of this application; however, the scope of protection of this application is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in this application, based on the technical solution and its improved concept, should be covered within the scope of protection of this application.
Claims
1. A method for detecting defects in the skin of an aero-engine, characterized in that, include: Acquire images of the aircraft engine skin; The image is preprocessed to remove image noise; A model for detecting defects in aero-engine skin is constructed: Feature parameters of the aero-engine skin surface under various aero-engine defects are extracted using a pre-trained neural network, and a set of rules for aero-engine skin defect feature parameters is constructed. Several defect classification sets are then trained using the masked scoring R-CNN algorithm, forming an aero-engine skin defect knowledge base. The neural network incorporates a convolutional block attention module and a feature fusion module. The convolutional block attention module includes a channel attention submodule and a spatial attention submodule. The channel attention submodule is used to mine the inter-channel relationships of features and filter out feature maps with strong discriminative power. The spatial attention submodule focuses on the position of image information parts and constructs a set of rules for aero-engine skin defect feature parameters. The preprocessed image to be tested is input into the aero-engine skin defect detection model to output aero-engine defect status information; The aircraft engine skin defect detection model includes: The feature parameter extraction module is used to extract feature parameters of the aero-engine skin surface under various aero-engine defects through a pre-trained neural network, and to construct a rule set of aero-engine skin defect feature parameters; The mask scoring R-CNN module is used for region candidate network processing, generating suggestions describing regions that may contain targets, and performing multi-class classification, candidate box regression, and introducing a fully convolutional neural network to generate masks for regions that may contain targets. Output module, used to output aircraft engine defects; The mask scoring R-CNN module includes: A region candidate network is used to generate suggestions describing regions that may contain the target. The classifier header is used to regress bounding box coordinates and class associations. Masking tool, used for binary mask calculation; The mask-iou header is used for mask scoring evaluation; The classifier head includes four convolutional layers and one fully connected layer, and the loss function is: ； in, This is a loss in the RPN classification task. The loss function used in masked R-CNN is the cross-entropy loss function. and These are the regression loss functions for RPN and Mask R-CNN, respectively, using a smoothed L1 loss function. It is the Mask-IoU regression loss, using the L2 loss function.
2. The method for detecting defects in the skin of an aero-engine according to claim 1, characterized in that, The preprocessing method uses median filtering to remove image noise.
3. A system for identifying defects in the skin of an aircraft engine, characterized in that, include: The image acquisition module is used to acquire images of the aircraft engine skin; A preprocessing module is used to preprocess the image to remove image noise; The model building module is used to construct an aero-engine skin defect detection model. It extracts feature parameters of the aero-engine skin surface under various aero-engine defects through a pre-trained neural network, constructs a rule set of aero-engine skin defect feature parameters, and then trains several defect classification sets using the masked scoring R-CNN algorithm to form an aero-engine skin defect knowledge base. The neural network includes a convolutional block attention module and a feature fusion module. The convolutional block attention module includes a channel attention submodule and a spatial attention submodule. The channel attention submodule is used to mine the inter-channel relationships of features and filter out feature maps with strong discriminative power. The spatial attention submodule focuses on the position of image information parts and constructs a rule set of aero-engine skin defect feature parameters. The output module outputs the defect status information of the aero-engine; The aircraft engine skin defect detection model includes: The feature parameter extraction module is used to extract feature parameters of the aero-engine skin surface under various aero-engine defects through a pre-trained neural network, and to construct a rule set of aero-engine skin defect feature parameters. The neural network includes a convolutional block attention module and a feature fusion module. The mask scoring R-CNN module is used for region candidate network processing, generating suggestions describing regions that may contain targets, and performing multi-class classification, candidate box regression, and introducing a fully convolutional neural network to generate masks for regions that may contain targets. Output module, used to output aircraft engine defects; The mask scoring R-CNN module includes: A region candidate network is used to generate suggestions describing regions that may contain the target. The classifier header is used to regress bounding box coordinates and class associations. Masking tool, used for binary mask calculation; The mask-iou header is used for mask scoring evaluation; The classifier head includes four convolutional layers and one fully connected layer, and the loss function is: ； in, This is a loss in the RPN classification task. The loss function used in masked R-CNN is the cross-entropy loss function. and These are the regression loss functions for RPN and Mask R-CNN, respectively, using a smoothed L1 loss function. It is the Mask-IoU regression loss, using the L2 loss function.