Rapid intelligent identification method and system for aviation assembly on-machine operation tool

Abstract

The invention belongs to the technical field of aeronautical manufacturing and assembly detection, and relates to a rapid intelligent identification method and system for an aeronautical assembly on-machine operation tool. The method comprises the steps that an aviation assembly on-machine operation tool image database is constructed, a tool recognition-oriented instance segmentation neural network model is designed, and the neural network model can recognize tool type feature information of on-machine operation tool images; training the instance segmentation neural network model by using a computer operation tool image database to obtain a primary computer operation tool recognition model, and processing the primary computer operation tool recognition model through a neural network distillation algorithm and a single-step rapid pruning algorithm to obtain a lightweight computer operation tool recognition model; the method comprises the following steps: acquiring a machine operation tool image in real time, performing instance segmentation on various tools in the image by utilizing a light-weight machine operation tool identification model, and marking various tools on the image by utilizing different colors to obtain a marked machine operation tool image.

Description

Technical Field

[0001] This application belongs to the field of aviation manufacturing and assembly inspection technology, and specifically relates to a rapid intelligent identification method and system for aviation assembly machine operation tools. Background Technology

[0002] In modern aircraft assembly, a large number of specialized tools are required for onboard operations, such as electric drills, torque wrenches, and measuring instruments. These tools are diverse, numerous, and frequently used. Incomplete or lost tools during assembly can severely impact production pace and product quality, and even jeopardize aircraft safety and reliability. Therefore, the automated identification and tracking of onboard tools is crucial for ensuring standardized operations and improving assembly efficiency.

[0003] Currently, the management of on-machine operation tools still mainly relies on manual inspection and manual recording, which has problems such as low efficiency, strong subjectivity, and easy omissions, making it difficult to meet the needs of high-speed, high-consistency, and high-reliability aerospace manufacturing. Summary of the Invention

[0004] Purpose of the invention: To provide a rapid and intelligent identification method and system for aircraft assembly tools, in order to solve the problems of low efficiency, high error rate and lack of real-time feedback in the current manual identification of tools.

[0005] Technical solution: A method for rapid and intelligent identification of tools used in aircraft assembly operations, comprising: Step 1: Construct an image database of aircraft assembly and onboard operation tools; Step 2: Based on the image database of the tools used in the machine operation, design an instance segmentation neural network model for tool recognition. The neural network model can identify the tool type feature information of the machine operation tool images. Step 3: Using the image database of the hands-on operation tools, train the instance segmentation neural network model to obtain a primary hands-on operation tool recognition model. Process the primary hands-on operation tool recognition model with the neural network distillation algorithm and the single-step fast pruning algorithm to obtain a lightweight hands-on operation tool recognition model. Step 4: Real-time acquisition of images of the on-machine operation tools; use a lightweight on-machine operation tool recognition model to segment various tools in the images; and mark each type of tool with different colors on the images to obtain the marked on-machine operation tool images.

[0006] Furthermore, in step one, the construction of the image database of aircraft assembly operation tools includes: High-definition industrial cameras are used to photograph the on-machine operations of various specialties being inspected, resulting in a predetermined number of images of the on-machine operation tools; The images of the hands-on tools are segmented and labeled at the instance level to generate corresponding label documents, wherein each image of the hands-on tools corresponds to one label document; A database of computer operation tool images is generated based on the images of the computer operation tools and the label documents.

[0007] Furthermore, the label document includes the name of the tool image, the type of each tool area, and coordinate information.

[0008] Furthermore, in step two, the on-machine operation tool recognition model algorithm is designed with a self-attention mechanism module based on the YOLOv8-seg convolutional neural network architecture, wherein... The design method of the self-attention mechanism module is as follows: Given input features , It belongs to the symbol. For real numbers, C For feature depth, H For feature height, W The feature width; Input features After depthwise separable convolution W dw1 Computation, in spatial dimensions H × W The upper part was reshaped to a size of 1× C of H × W There are eigenvectors, denoted as eigenvectors. I ,in, H × W Each of the eigenvectors represents feature information of a specific spatial location; Subsequently, features I Through a learnable weight matrix , , The transformation is performed to generate the tensor representations of the corresponding query Q, key K, and value V, which are calculated as follows:

[0009]

[0010]

[0011] The obtained Q, K, and V tensors are used to calculate attention, and the attention matrix is ​​output. and feature map The calculation is as follows:

[0012]

[0013] In the formula, It is a normalization function used to calculate the attention weight matrix. ; feature map After passing through a size of ( C , H × W Depth-separable convolutions of type 1, 1, and 1) Wdw 2. Calculation and reshaping to obtain features Through the With feature map Perform element-wise addition to obtain the corrected feature map. This optimizes the original feature representation.

[0014] Furthermore, in step three, a neural network distillation algorithm is used to process the primary on-machine operation tool recognition model, specifically as follows: Attention distillation and channel-space distillation losses are set for the basic computer lab assignment tool recognition model during model training. Specifically, the basic computer lab assignment tool recognition model is used to guide the knowledge distillation process to obtain the distilled student model. The feature attention distillation loss function is calculated as follows:

[0015] In the formula, N represents the sample size. and These represent the attention matrices of the teacher model and the student model, respectively. This represents the Euclidean distance, i.e., the L2 norm. The formula for calculating the feature channel-spatial distillation loss function is as follows:

[0016] In the formula, and The teacher model and student model respectively represent the first two models in the second stage. i The spatial location and the first c Feature values ​​on each channel and These are obtained by performing max pooling operations along the channel axis on the output features of the teacher model and the student model, respectively. This represents the number of output channels in the teacher network, and the output features of the student network are represented by a size of [size missing]. The convolution operation performs an output dimension transformation, where... This indicates the number of output channels in the student network.

[0017] Furthermore, in step three, a single-step fast pruning algorithm is used to process the distilled student model, specifically as follows: The distilled student model undergoes dynamic sparse training and single-step pruning. The dynamic sparse training process involves applying regularization techniques to the backbone and neck structures of the distilled student model. This is done by identifying the feature channels with the lowest activation levels based on the L1 norm. After selecting a channel, the L2 norm of the feature channel with the lowest activation level is calculated and added as a regularization term to the training loss function. The dynamic sparse regularization term is defined as follows:

[0018] In the formula, This indicates the global mean pooling operation. Indicates the first i The selected feature maps in the layer are subject to sparse regularization constraints. The weights of the regularization terms associated with each layer in the network are represented. After applying sparse regularization, a structured pruning process is performed to remove the weight channels associated with the pruned feature channels, thus obtaining the final lightweight on-the-job tool recognition model.

[0019] Furthermore, it also includes: Step 5: Count the number of each type of tool, and feed the statistical results and the marked images of the onboard operation tools back to the boarding operation data management system for storage and display.

[0020] A rapid intelligent identification system for aircraft assembly and loading tools, used to perform the above method, includes: This includes high-definition industrial cameras, model inference servers, on-machine tool placement platforms, image preprocessing modules, and on-machine operation data management systems. The high-definition industrial camera captures images of the tools placed on the work platform in real time and transmits these images to a model inference server via a wired network. The model inference server has a built-in image preprocessing module and a lightweight work tool recognition model. The image preprocessing module is used to crop, denoise, and enhance the images to improve recognition accuracy. The lightweight work tool recognition model can identify and statistically analyze various professional tools to obtain statistical results. The boarding operation data management system can record and manage the statistical results of the work tools.

[0021] Beneficial effects: This application discloses a rapid and intelligent identification method for on-machine tools used in aviation assembly. First, high-definition industrial cameras are used to capture images of the on-machine tools, and instance-level segmentation and annotation are performed on various tools within the images to establish an image database of on-machine tools. An instance segmentation neural network model for tool recognition is then constructed. Next, the deep neural network model is trained using the on-machine tool image database until the training loss value converges and the test accuracy stabilizes, resulting in an on-machine tool recognition model. The on-machine tool recognition model is then optimized through distillation and pruning to obtain a final lightweight on-machine tool recognition model. During operation, real-time acquisition of images from the work site allows for automatic instance segmentation and quantity counting of various types of on-machine tools using the lightweight recognition model. This enables rapid batch analysis and processing of work site images, significantly improving the efficiency and accuracy of tool identification and management. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart of a rapid intelligent identification method for aircraft assembly and loading tools according to one embodiment of this application; Figure 2 This is a schematic diagram of a self-attention module design according to one embodiment of this application; Figure 3 This is a schematic diagram of the neural network distillation algorithm design according to one embodiment of this application; Figure 4 This is a schematic diagram of a rapid intelligent identification system for aircraft assembly and loading tools according to one embodiment of this application.

[0024] in: 100 - Module for constructing datasets for computer-based assignment tools; 200 - Module for constructing recognition models for computer-based assignment tools; 300 - Module for model training and lightweighting; 400 - Module for model inference and recognition; 500 - Module for statistical analysis and management of recognition results. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0026] The features and illustrative embodiments of various aspects of the present invention will now be described in detail. Numerous specific details are set forth in the following detailed description to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention may be practiced without requiring some of these specific details. The following description of embodiments is merely intended to provide a better understanding of the invention by illustrating examples of the invention. The invention is by no means limited to any specific setups and methods set forth below, but covers any improvements, substitutions, and modifications to structures, methods, and devices without departing from the spirit of the invention. Well-known structures and techniques are not shown in the drawings and the following description to avoid unnecessarily obscuring the invention.

[0027] In the description of this invention, it should be noted that the directions or positional relationships indicated by terms such as "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer" are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing and simplifying the invention, and should not be construed as limiting the invention. Furthermore, the use of ordinal numbers (e.g., "first and second," etc.) is for distinguishing objects and is not limited to this order, and should not be construed as indicating or implying relative importance.

[0028] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly, encompassing both direct connection and indirect connection via an intermediate medium. Those skilled in the art can understand the specific meaning of these terms in this invention based on the specific circumstances.

[0029] It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other, and the various embodiments can be referenced and cited in each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0030] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0031] With the development of computer vision and artificial intelligence technologies, image recognition-based tool detection methods are gradually becoming an effective means to improve the level of operation management.

[0032] By acquiring images of tools on-site using high-definition industrial cameras and combining them with an instance segmentation neural network model to accurately identify and classify the types and locations of tools in the images, it is possible to effectively achieve rapid tool statistics, status monitoring, and image visualization annotation, providing technical support for digital assembly operations and intelligent tool management.

[0033] This application belongs to the field of aviation manufacturing and assembly inspection technology, and specifically relates to a rapid intelligent identification method and system for aviation assembly machine operation tools. The method includes: Step 1: Using a high-definition industrial camera to acquire images of various types of professional tools in the machine operation scene, and performing instance-level segmentation and classification labeling on the tools in the images to generate corresponding label documents and construct an image database of aviation assembly machine operation tools; Step 2: Designing an instance segmentation neural network model for tool recognition, introducing a self-attention mechanism module based on the YOLOv8-seg architecture to optimize feature representation; Step 3: Training the instance segmentation neural network model using the constructed image database to obtain a machine operation tool recognition model; and designing a neural network distillation algorithm and a single-step fast pruning algorithm for the model to optimize the model structure and obtain a lightweight machine operation tool recognition model; Step 4: Real-time acquisition of tool images from the work site, using the lightweight recognition model to perform rapid instance segmentation and classification recognition of the tools in the images, and visually marking each type of tool in the images with different colors to obtain labeled machine operation tool images; Step 5: Counting the number of each type of tool, and feeding the statistical results and labeled images back to the boarding operation data management system for storage and display. This application enables rapid and accurate identification and statistics of on-machine operation tools, improves detection efficiency and accuracy, reduces human error, and provides intelligent management support for aviation assembly operations.

[0034] The following is in conjunction with the appendix Figures 1 to 4 This application will be described in further detail.

[0035] The first aspect of this application provides a method for rapid and intelligent identification of tools used in aircraft assembly operations, such as... Figure 1 As shown, it includes the following steps: Step 1: Construct an image database of aircraft assembly and onboard operation tools; Step 2: Design an instance segmentation neural network model for tool recognition based on the image database of on-machine operation tools. The neural network model can identify the tool type feature information of the on-machine operation tool images. Step 3: Using the image database of the hands-on operation tools, train the instance segmentation neural network model to obtain a primary hands-on operation tool recognition model. Process the primary hands-on operation tool recognition model with the neural network distillation algorithm and the single-step fast pruning algorithm to obtain a lightweight hands-on operation tool recognition model. Step 4: Real-time acquisition of images of the on-machine operation tools; use a lightweight on-machine operation tool recognition model to segment various tools in the images; and mark each type of tool with different colors on the images to obtain the marked on-machine operation tool images.

[0036] In a preferred embodiment of this application, it further includes: Step 5: Count the number of each type of tool, and feed the statistical results and the marked images of the onboard operation tools back to the boarding operation data management system for storage and display.

[0037] The rapid and intelligent identification method for aircraft assembly machine operation tools in this application includes, in step one, constructing an image database of aircraft assembly machine operation tools, including: High-definition industrial cameras are used to photograph the on-machine operations of various specialties being inspected, resulting in a predetermined number of images of the on-machine operation tools; Perform instance-level segmentation and annotation on the images of the computer tools, and generate corresponding label documents. Each image of the computer tools corresponds to one label document. A database of images of computer operation tools is generated based on the images of the tools and the label documents.

[0038] In a preferred embodiment of this application, before taking pictures of the tools used on the machine, the tools are placed in a designated area; a high-definition industrial camera takes pictures of the tools in that area each time; during image analysis, the statistical boundary range of the tool images is set according to the designated area; when performing instance-level segmentation and labeling of each type of tool in each tool image, a corresponding JSON-formatted tag document is generated. The tool image database includes tool images and corresponding tag documents, and the tag document includes the name of the tool image, the coordinates of the tool area, and the type information.

[0039] The rapid intelligent identification method for aircraft assembly tools in this application, in step two, involves designing a self-attention mechanism module algorithm based on the YOLOv8-seg convolutional neural network architecture to improve the image recognition and segmentation effect of the tools. In this embodiment, as shown... Figure 2 As shown, the self-attention mechanism module unit design architecture.

[0040] The design method for the self-attention mechanism module is as follows: Given input features , It belongs to the symbol. For real numbers, C For feature depth, H For feature height, W The feature width; input features After depthwise separable convolution Wdw 1. Calculation, in spatial dimension H × W The upper part was reshaped to a size of 1× C of H × W There are eigenvectors, denoted as eigenvectors. I .in, H × W Each of the eigenvectors represents feature information at a specific spatial location. Subsequently, the features... I Through a learnable weight matrix , , The transformation is performed to generate tensor representations of the corresponding query (Query, Q), key (Key, K), and value (Value, V), which are calculated as follows:

[0041]

[0042]

[0043] The obtained Q, K, and V tensors are used to calculate attention, and the attention matrix is ​​output. and feature map The calculation is as follows:

[0044]

[0045] In the formula, It is a normalization function used to calculate the attention weight matrix. Feature map After passing through a size of ( C , H × W Depth-separable convolutions of type 1, 1, and 1) Wdw 2. Calculation and reshaping to obtain features Through the With feature map Perform element-wise addition to obtain the corrected feature map. This optimizes the original feature representation.

[0046] The rapid intelligent identification method for aircraft assembly tools disclosed in this application, in step three, employs a neural network distillation algorithm to process the primary tool identification model. This reduces the computational load of the tool identification model while ensuring effective tool image recognition and segmentation. In this embodiment, as shown... Figure 3 As shown, the design architecture of the model distillation unit is presented.

[0047] The design method of the model distillation algorithm is as follows: Attention distillation and channel-space distillation losses are set for the basic computer lab assignment tool recognition model during model training. Specifically, the basic computer lab assignment tool recognition model is used to guide the knowledge distillation process to obtain the distilled student model. The feature attention distillation loss function is calculated as follows:

[0048] In the formula, N represents the sample size. and These represent the attention matrices of the teacher model and the student model, respectively. This represents the Euclidean distance (L2 norm). The formula for calculating the feature channel-spatial distillation loss function is:

[0049] In the formula, and The teacher model and student model respectively represent the first two models in the second stage. i The spatial location and the first c Feature values ​​on each channel and These are obtained by performing max pooling operations along the channel axis on the output features of the teacher model and the student model, respectively. This represents the number of output channels in the teacher network, and the output features of the student network are represented by a size of [size missing]. The convolution operation performs an output dimension transformation, where... This indicates the number of output channels in the student network.

[0050] The rapid intelligent recognition method for aircraft assembly tools in this application, in step three, employs a single-step rapid pruning algorithm to process the distilled student model, achieving rapid single-step model size compression, further reducing the computational load of the tool recognition model, and improving the efficiency of tool image recognition and analysis. In this embodiment, as... Figure 3 As shown, the self-attention mechanism module unit design architecture.

[0051] The design method of the single-step fast pruning algorithm for the model is as follows: The distilled student model undergoes dynamic sparse training and single-step pruning. The dynamic sparse training process involves applying regularization techniques to the backbone and neck structures of the distilled student model. This is done by identifying the feature channels with the lowest activation levels based on the L1 norm. After selecting a channel, the L2 norm of the feature channel with the lowest activation level is calculated and added as a regularization term to the training loss function. The dynamic sparse regularization term is defined as follows:

[0052] In the formula, This indicates the global mean pooling operation. Indicates the first i The selected feature maps in the layer are subject to sparse regularization constraints. This represents the weight of the regularization term associated with each layer in the network. After applying sparse regularization, a structured pruning process is performed to remove the weight channels associated with the pruned feature channels, resulting in the final lightweight on-the-job tool recognition model.

[0053] The rapid and intelligent identification method for aircraft assembly tools disclosed in this application, in step four, utilizes a lightweight tool identification model to segment real-time acquired tool images and marks each type of tool with different colors on the images, obtaining marked tool images. This avoids the drawback of manually counting all tools, reducing labor intensity and improving assembly efficiency. Finally, in step five, the quantity of each type of tool is counted, and the statistical results and marked tool images are fed back to the boarding operation data management system for storage and display, enabling tool record verification and backup.

[0054] A rapid intelligent identification system for aircraft assembly tools includes: a high-definition industrial camera, a model inference server, a tool placement platform, an image preprocessing module, and an aircraft loading data management system. The high-definition industrial camera captures images of the tools placed on the work platform in real time and transmits these images to a model inference server via a wired network. The model inference server has a built-in image preprocessing module and a lightweight work tool recognition model. The image preprocessing module is used to crop, denoise, and enhance the images to improve recognition accuracy. The lightweight work tool recognition model can identify and statistically analyze various professional tools to obtain statistical results. The boarding operation data management system can record and manage the statistical results of the work tools.

[0055] The rapid and intelligent identification method for aviation assembly tools disclosed in this application first uses a high-definition industrial camera to acquire images of the tools, and performs instance-level segmentation and annotation on various tools in the images to establish an image database of the tools. Then, it builds an instance segmentation neural network model for tool recognition. Next, it trains the deep neural network model using the tool image database until the training loss value converges and the test accuracy stabilizes, obtaining the tool recognition model. The tool recognition model is then distilled and pruned to optimize it, resulting in a final lightweight tool recognition model. During operation, by acquiring real-time images of the work site, the lightweight recognition model can automatically segment and count the number of various types of tools, achieving automated detection of assembly work, thereby improving assembly work efficiency and realizing intelligent management and control.

[0056] The rapid intelligent identification method for aviation assembly tools in this application is based on the algorithm design of self-attention module, distillation module, pruning module and target sample recognition and segmentation. It can automatically learn and extract tool features from the tool images of the assembly tools, realize rapid batch analysis and processing of on-site images, and greatly improve the efficiency and accuracy of tool identification and management.

[0057] Based on the aforementioned method for rapid and intelligent identification of aircraft assembly tools, a second aspect of this application provides a system for rapid and intelligent identification of aircraft assembly tools, such as... Figure 4 As shown, it includes: The on-machine operation tool dataset construction module 100 is used to construct an image database of aviation assembly on-machine tools; The "Machine Tool Recognition Model Construction Module 200" is used to design and implement an instance segmentation neural network model for tool recognition. The model training and lightweight module 300 is used to train a deep neural network model using an image database of on-the-job tools, and obtain a lightweight on-the-job tool recognition model by designing a neural network distillation algorithm and a single-step fast pruning algorithm. The model reasoning and recognition module 400 is used to acquire images of on-machine operation tools in real time, use a lightweight on-machine operation tool recognition model to segment various tools in the image, and mark each type of tool on the image with different colors to obtain the marked on-machine operation tool image; The identification result statistical analysis and management module 500 is used to count the number of tools of each type and to feed the statistical results and the marked images of the onboard operation tools back to the boarding operation data management system for storage and display.

[0058] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for rapid and intelligent identification of tools used in aircraft assembly operations, characterized in that, include: Step 1: Construct an image database of aircraft assembly and onboard operation tools; Step 2: Based on the image database of the tools used in the machine operation, design an instance segmentation neural network model for tool recognition. The neural network model can identify the tool type feature information of the machine operation tool images. Step 3: Using the image database of the hands-on operation tools, train the instance segmentation neural network model to obtain a primary hands-on operation tool recognition model. Process the primary hands-on operation tool recognition model with the neural network distillation algorithm and the single-step fast pruning algorithm to obtain a lightweight hands-on operation tool recognition model. Step 4: Real-time acquisition of images of the on-machine operation tools; use a lightweight on-machine operation tool recognition model to segment various tools in the images; and mark each type of tool with different colors on the images to obtain the marked on-machine operation tool images.

2. The rapid intelligent identification method for aircraft assembly tools according to claim 1, characterized in that, Step one, the construction of the image database of aircraft assembly and onboard operation tools, includes: High-definition industrial cameras are used to photograph the on-machine operations of various specialties being inspected, resulting in a predetermined number of images of the on-machine operation tools; The images of the hands-on tools are segmented and labeled at the instance level to generate corresponding label documents, wherein each image of the hands-on tools corresponds to one label document; A database of computer operation tool images is generated based on the images of the computer operation tools and the label documents.

3. The rapid intelligent identification method for aircraft assembly tools according to claim 2, characterized in that, The tag document includes the name of the tool image, the type of each tool area, and coordinate information.

4. The rapid intelligent identification method for aircraft assembly tools according to claim 3, characterized in that, In step two, the on-machine operation tool recognition model algorithm is designed with a self-attention mechanism module based on the YOLOv8-seg convolutional neural network architecture. The design method of the self-attention mechanism module is as follows: Given input features , It belongs to the symbol. For real numbers, C For feature depth, H For feature height, W The feature width; Input features After depthwise separable convolution W dw1 Computation, in spatial dimensions H × W The upper part was reshaped to a size of 1× C of H × W There are eigenvectors, denoted as eigenvectors. I ,in, H × W Each of the eigenvectors represents feature information of a specific spatial location; Subsequently, features I Through a learnable weight matrix , , The transformation is performed to generate the tensor representations of the corresponding query Q, key K, and value V, which are calculated as follows: The obtained Q, K, and V tensors are used to calculate attention, and the attention matrix is ​​output. and feature map The calculation is as follows: In the formula, It is a normalization function used to calculate the attention weight matrix. ; feature map After passing through a size of ( C , H × W Depth-separable convolutions of (1, 1,) Wdw 2. Calculation and reshaping to obtain features Through the With feature map Perform element-wise addition to obtain the corrected feature map. This optimizes the original feature representation.

5. The rapid intelligent identification method for aircraft assembly tools according to claim 4, characterized in that, In step three, a neural network distillation algorithm is used to process the primary on-machine operation tool recognition model, specifically as follows: Attention distillation and channel-space distillation losses are set for the basic computer lab assignment tool recognition model during model training. Specifically, the basic computer lab assignment tool recognition model is used to guide the knowledge distillation process to obtain the distilled student model. The feature attention distillation loss function is calculated as follows: In the formula, N represents the sample size. and These represent the attention matrices of the teacher model and the student model, respectively. This represents the Euclidean distance, i.e., the L2 norm. The formula for calculating the feature channel-spatial distillation loss function is as follows: In the formula, and The teacher model and student model respectively represent the first two models in the second stage. i The spatial location and the first c Feature values ​​on each channel and These are obtained by performing max pooling operations along the channel axis on the output features of the teacher model and the student model, respectively. This represents the number of output channels in the teacher network, and the output features of the student network are represented by a size of [size missing]. The convolution operation performs an output dimension transformation, where... This indicates the number of output channels in the student network.

6. The rapid intelligent identification method for aircraft assembly tools according to claim 5, characterized in that, In step three, a single-step fast pruning algorithm is used to process the distilled student model, specifically as follows: The distilled student model undergoes dynamic sparse training and single-step pruning. The dynamic sparse training process involves applying regularization techniques to the backbone and neck structures of the distilled student model. This is done by identifying the feature channels with the lowest activation levels based on the L1 norm. After selecting a channel, the L2 norm of the feature channel with the lowest activation level is calculated and added as a regularization term to the training loss function. The dynamic sparse regularization term is defined as follows: In the formula, This indicates the global mean pooling operation. Indicates the first i The selected feature maps in the layer are subject to sparse regularization constraints. The weights of the regularization terms associated with each layer in the network are represented. After applying sparse regularization, a structured pruning process is performed to remove the weight channels associated with the pruned feature channels, thus obtaining the final lightweight on-the-job tool recognition model.

7. The rapid intelligent identification method for aircraft assembly tools according to claim 6, characterized in that, Also includes: Step 5: Count the number of each type of tool, and feed the statistical results and the marked images of the onboard operation tools back to the boarding operation data management system for storage and display.

8. A rapid intelligent identification system for aircraft assembly tools, used to execute the method as described in any one of claims 1-7, characterized in that, include: This includes high-definition industrial cameras, model inference servers, on-machine tool placement platforms, image preprocessing modules, and on-machine operation data management systems. The high-definition industrial camera captures images of the tools placed on the work platform in real time and transmits these images to a model inference server via a wired network. The model inference server has a built-in image preprocessing module and a lightweight work tool recognition model. The image preprocessing module is used to crop, denoise, and enhance the images to improve recognition accuracy. The lightweight work tool recognition model can identify and statistically analyze various professional tools to obtain statistical results. The boarding operation data management system can record and manage the statistical results of the work tools.

Images