Research On The Recognition And Extraction Of Closed Contours Of Part Diagrams Based On Neural Networks

At present,the technology of obtaining the 2D part diagram of a part by projecting the3 D solid model of a mechanical part is quite mature,but the reconstruction of the 3D model of a part by 2D image of the part and related information is currently done by manual operation,which is tedious and inefficient and cannot meet the needs of the rapid development of intelligent manufacturing and digital integrated manufacturing technology.In this project,we take common shaft,disk,fork and box parts as the objects and carry out the research on the closed contour of the parts diagram.Identifying and extracting the closed contours of mechanical parts engineering drawings lays the foundation for 3D model reconstruction and allows the extracted closed contours to be cut off the corresponding parts on sheet metal materials to realize the problem of undercutting simple parts.Aiming at the problems of low recognition and extraction efficiency,low precision and single recognition and extraction object of traditional methods,based on neural networks the project of the recognition and extraction of closed contours of part diagrams is dedicated to solve the problem of low efficiency in the process of recognition and extraction of closed contours in mechanical parts diagram,based on which a neural network model is proposed for the recognition and extraction of closed contours in mechanical parts diagram.Based on neural networks the project of the recognition and extraction of closed contours of part diagrams is implemented in the Tensor Flow deep learning framework based on Python language.The main contents and results are as follows:(1)Machine parts data set preparation and processing.Firstly,we use Auto CAD drawing software to draw a complete mechanical parts diagram,because this topic focuses on the closed contour of the parts diagram,firstly,we close the layers that are not needed for this study,and only keep the thick solid line layer;secondly,we use labelme software to mark and name the closed contour of different shapes,and generate the corresponding semantic segmentation diagram;finally,in order to prevent the overfitting phenomenon to expand the dataset,we use flip and crop to expand the dataset.Finally,to prevent overfitting,the dataset is expanded using data augmentation tools such as flip and crop.(2)Aiming at the problems of low efficiency and single graphic objects in the process of identifying and extracting mechanical parts drawings using traditional methods,a recognition and extraction model based on convolutional neural networks applied to mechanical parts drawings is proposed.First,the selected simple mechanical axis graphics are identified and extracted,and the data augmentation method is used to expand it to solve the problem of insufficient data sets.The semantic segmentation network FCNs(Fully Convolutional Networks),Seg Net and U-Net is trained and tested on the shaft parts dataset.Through the verification and analysis of the experimental results with a number of evaluation indicators,the traditional U-Net model has better performance than FCNs and Seg Net in the recognition and extraction of the closed contour of mechanical parts.In addition,the U-Net network model has better performance in the segmentation of mechanical graphics has better generalization,which preliminarily verifies the feasibility of using U-Net to identify and extract the closed contours of mechanical part drawings.(3)The recognition extraction accuracy of traditional U-Net model for mechanical parts graphics is significantly better than FCNs and Seg Net models,but the extracted closed contours of some parts graphics show pixel classification errors and missed detection.To address this problem,this study proposes an improved model based on U-Net for the recognition and extraction of closed contours of mechanical part graphics,namely,adding CBAM(Convolutional Block Attention Module)attention mechanism after upsampling and MASPP(Mean Atrous Spatial Pyramid Pooling)before upsampling to the U-Net decoder.(Spatial Pyramid Pooling)pyramid module before upsampling to improve the segmentation performance of its model.The U-Net network with the CBAM attention mechanism and MASPP pyramid module has significantly improved the phenomenon of missed detection and pixel classification error,and the contour details are more finely identified and extracted,and the overall segmentation accuracy is significantly improved compared with the original U-Net network and the U-Net network with only one module added.The overall segmentation accuracy is significantly improved.Compared with traditional methods,the semantic segmentation model based on convolutional neural network has higher efficiency and universality in the task of recognition and extraction of closed contours of mechanical parts.Compared with the original U-Net network,the improved U-Net model that integrates the CBAM attention mechanism and the MASPP pyramid structure proposed in this study can better realize the identification and extraction of the closed contour of the mechanical part diagram.