Typical mechanical part identification method oriented to body intelligence

Abstract

The invention discloses a typical mechanical part identification method oriented to body intelligence, and the method comprises the steps: S1, data set preparation: collecting and marking a large number of mechanical part images, and constructing a training data set; s2, model training: training a training data set by using an improved Faster R-CNN network, optimizing hyper-parameters, and ensuring optimal model performance; and S3, model evaluation: through comparison with an existing identification technology, evaluating the effect of the improved model in practical application, and carrying out iterative optimization. According to the invention, an improved Faster R-CNN backbone network is provided for typical mechanical part recognition in a body-oriented intelligent scene, so that the recognition speed and accuracy of mechanical parts are improved, and the requirements of aerospace and intelligent manufacturing fields for efficient recognition technologies are met.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and artificial intelligence, and in particular to a typical mechanical parts recognition method oriented to embodied intelligence. Background Art

[0002] With the development of embodied intelligence, traditional image recognition methods face multiple challenges in mechanical parts recognition. Through perception and interaction, embodied intelligence can help robots adaptively process and recognize different objects in complex environments, and is particularly suitable for industrial scenarios that require high precision and robustness. However, the existing mechanical parts recognition technology has the following bottlenecks in combining embodied intelligence:

[0003] (1) Traditional image processing technology relies on feature extraction and matching algorithms, which often cannot guarantee high accuracy in complex backgrounds or under different lighting conditions. The shape and color of mechanical parts may change due to environmental factors, resulting in misidentification or missed identification, which cannot meet the high accuracy requirements of industrial applications.

[0004] (2) Existing recognition methods usually take a long time to perform feature calculation and pattern matching when processing complex images. Especially in large-scale production environments, this delay will lead to reduced production efficiency and cannot meet the needs of real-time recognition.

[0005] (3) Traditional mechanical parts recognition methods perform poorly in the case of image noise, deformation, and occlusion, and lack sufficient robustness. This makes it easy for the system to fail in practical applications when the parts are in a non-ideal state, affecting the overall production process. Summary of the invention

[0006] In view of this, the purpose of the present invention is to provide a typical mechanical parts recognition method for embodied intelligence. Through the improved FasterR-CNN backbone network, the recognition speed and accuracy of mechanical parts can be improved to meet the needs of aerospace and intelligent manufacturing fields for efficient recognition technology.

[0007] The present invention solves the technical problem by adopting the following technical solution:

[0008] A typical mechanical parts recognition method for embodied intelligence includes the following steps:

[0009] Step S1, data set preparation: collect and annotate a large number of mechanical parts images to construct a training data set;

[0010] Step S2, model training: use the improved Faster R-CNN network to train the training data set and optimize the hyperparameters to ensure the best model performance;

[0011] Step S3, model evaluation: by comparing with existing recognition technologies, evaluate the effect of the improved model in practical applications and perform iterative optimization.

[0012] Furthermore, the improved Faster R-CNN network structure includes CBAM module, RAN module and ResNeXt residual block;

[0013] The CBAM module applies attention mechanisms on channels and spaces, enabling the robot to automatically adjust the degree of attention paid to key features and achieve stable recognition in complex backgrounds;

[0014] The RAN module uses residual connections to improve information flow and reduce the gradient vanishing problem, so that the network can still maintain good learning ability when it is deeper;

[0015] The ResNeXt residual block extracts part features faster through action perception feedback of embodied intelligence and uses grouped convolution to accelerate calculations, supporting robots to quickly adapt to different tasks in diverse production environments.

[0016] Furthermore, the CBAM module adopts the module design idea of ​​channel attention mechanism + spatial attention mechanism. By calculating the importance of each channel, it dynamically adjusts the weight of the feature map, highlights useful features, suppresses redundant information, and generates a spatial attention map based on channel information, so that the network can focus on the characteristics of specific areas and improve the response ability to key parts.

[0017] Furthermore, hyperparameters include learning rate, momentum parameter, number of parameters, batch size, training rounds, optimizer, and initial weights.

[0018] The present invention discloses a typical mechanical parts identification method for embodied intelligence, which has the following beneficial effects:

[0019] 1. Fast recognition speed: The improved Faster R-CNN network can quickly process input data in a real-time environment, significantly improving recognition efficiency.

[0020] 2. High recognition accuracy: Through multi-level feature extraction and fusion, the model shows high recognition accuracy in different scenarios.

[0021] 3. Wide applicability: This technology can be widely used in aerospace, intelligent manufacturing and other industries to improve production efficiency and product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flow chart of the method of the present invention;

[0023] Figure 2 A schematic diagram of the improved Faster R-CNN network structure of the present invention;

[0024] Figure 3 This is a schematic diagram of the CBAM module of the present invention;

[0025] Figure 4 It is a detailed implementation diagram of the CBAM module of the present invention;

[0026] Figure 5 This is a schematic diagram of the RAN module of the present invention;

[0027] Figure 6 It is a schematic diagram of the detailed implementation of the RAN module of the present invention;

[0028] Figure 7 Schematic diagram of the ResNeXt residual block of the present invention;

[0029] Figure 8 This is a schematic diagram of some typical mechanical parts recognition results of the present invention;

[0030] Fig. 9 Prepare schematic diagram for the inventive dataset. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] refer to Figure 1 The present invention discloses a typical mechanical parts recognition method for embodied intelligence, comprising the following steps:

[0033] Step S1, data set preparation: collect and annotate a large number of mechanical parts images, build a training data set, and refer to Fig. 9 ;

[0034] Step S2, model training: use the improved Faster R-CNN network to train the training data set and optimize the hyperparameters to ensure the best model performance;

[0035] Step S3, model evaluation: by comparing with existing recognition technologies, evaluate the effect of the improved model in practical applications and perform iterative optimization.

[0036] refer to Figure 2 ,The improved Faster R-CNN network structure includes CBAM module, RAN module and ResNeXt residual block;

[0037] refer to Figure 3 and Figure 4 , the CBAM module (convolutional block attention module) applies attention mechanism on channels and spaces, so that the robot can automatically adjust the degree of attention to key features and achieve stable recognition under complex backgrounds; the CBAM module adopts the module design idea of ​​channel attention mechanism + spatial attention mechanism, calculates the importance of each channel, dynamically adjusts the weight of the feature map, highlights useful features, suppresses redundant information, and generates a spatial attention map in combination with channel information, so that the network can focus on the features of specific areas and improve the responsiveness to key parts. Through fine-tuning of features, the system is more sensitive to dynamically changing scenes and adapts to the real-time recognition needs of industrial sites.

[0038] refer to Figure 5 and Figure 6 , the RAN module (residual attention network) uses residual connections to improve information flow and reduce the gradient vanishing problem, so that the network can still maintain good learning ability when it is deeper; by introducing the mask attention mechanism, it can adaptively select important features and suppress irrelevant features, thereby enhancing the expression ability of the model. This structure can maintain efficient calculation while extracting deep features, and is suitable for large-scale image processing.

[0039] refer to Figure 7 , ResNeXt residual blocks extract part features faster through embodied intelligent action perception feedback, and use grouped convolution (number of groups = number of input channels / radix, the default radix is ​​32) to accelerate calculations, supporting robots to quickly adapt to different tasks in diverse production environments. ResNeXt residual blocks can effectively improve computing efficiency while maintaining accuracy, and are suitable for industrial environments that require real-time response.

[0040] This paper combines the concept of embodied intelligence to improve the backbone network of Faster R-CNN, introduces new convolutional layers and feature extraction strategies, optimizes the network structure, and improves computing efficiency. A mechanical parts recognition system driven by embodied intelligence is constructed by combining the network structure of CBAM (Convolutional Block Attention Module) and RAN (Residual Attention Network), which enables robots to achieve efficient and accurate recognition in a dynamic production environment. Figure 8 (a) Figure 8 (b) Figure 8 (c) Figure 8 (d) Figure 8 (e) and Figure 8 (f).

[0041] To further optimize the technical solution, the hyperparameters include learning rate, momentum parameter, parameter amount, batch size, training rounds, optimizer, and initial weights, see Table 1.

[0042]

[0043]

[0044] Table 1

[0045] The present invention discloses a typical mechanical parts recognition method for embodied intelligence, which improves the accuracy of mechanical parts recognition by optimizing algorithms and feature extraction mechanisms; improves the recognition speed of mechanical parts recognition by optimizing network structures; improves the adaptability of the model to various interferences (such as illumination changes, partial occlusions, part deformations, etc.) to ensure stable operation in a changing working environment; and is designed to adapt to various types of mechanical parts to improve versatility in fields such as aerospace and intelligent manufacturing.

[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A typical mechanical parts recognition method for embodied intelligence, characterized in that: The steps include: Step S1, data set preparation: collect and annotate a large number of mechanical parts images to construct a training data set; Step S2, model training: use the improved Faster R-CNN network to train the training data set and optimize the hyperparameters to ensure the best model performance; Step S3, model evaluation: by comparing with existing recognition technologies, evaluate the effect of the improved model in practical applications and perform iterative optimization.

2. A typical mechanical parts identification method for embodied intelligence according to claim 1, characterized in that: The improved Faster R-CNN network structure includes CBAM module, RAN module and ResNeXt residual block; The CBAM module applies attention mechanisms on channels and spaces, enabling the robot to automatically adjust the degree of attention paid to key features and achieve stable recognition in complex backgrounds; The RAN module uses residual connections to improve information flow and reduce the gradient vanishing problem, so that the network can still maintain good learning ability when it is deeper; The ResNeXt residual block extracts part features faster through action perception feedback of embodied intelligence and uses grouped convolution to accelerate calculations, supporting robots to quickly adapt to different tasks in diverse production environments.

3. A typical mechanical parts identification method for embodied intelligence according to claim 2, characterized in that: The CBAM module adopts the module design idea of ​​channel attention mechanism + spatial attention mechanism. It calculates the importance of each channel, dynamically adjusts the weight of the feature map, highlights useful features, suppresses redundant information, and generates a spatial attention map in combination with channel information, so that the network can focus on the characteristics of specific areas and improve the responsiveness to key parts.

4. A typical mechanical parts identification method for embodied intelligence according to claim 3, characterized in that: Hyperparameters include learning rate, momentum parameter, number of parameters, batch size, number of training epochs, optimizer, and initial weights.

Images