Multi-view Change Detection And Pose Estimation Method For Mechanical Assembly Based On Deep Learning

The assembly process of mechanical products is characterized by many operation links and complex assembly process,which is easy to cause errors such as missing assembly and wrong assembly.In the assembly process of complex assembly parts,the quality and assembly efficiency of mechanical products will be affected if the new assembly parts are not checked in time.Therefore,this paper studied the deep learnbased multi-perspective change detection and pose estimation methods of mechanical assembly,and proposed assembly sequence monitoring method based on image change detection technology and assembly pose monitoring method based on pose estimation technology.The specific research contents are as follows:(1)An assembly multi-view change detection method based on feature extraction of deep image attention mechanism was proposed,which included semantic fusion network and multi-view change detection network.Semantic fusion network fuses semantic image and depth image to obtain fusion image,which enriches the diversity of image feature information.In this study,the self-attention mechanism and multi-scale feature fusion mechanism are applied to multi-perspective change detection network.The self-attention mechanism captures the global position dependence of feature information,and the multi-scale feature fusion mechanism effectively integrates highdimensional semantic information and low-dimensional spatial information.In addition,this research creates the mechanical assembly body to synthesize image change detection data set,and multiple points of view and to carry out the experimental study,the experimental results show that the proposed feature extraction based on depth image attention mechanism is the assembly of multiple points of view change detection method on the data set to 96.9%,the comprehensive index of F1 can realize the mechanical assembly body depth image change detection multiple points of view.(2)An assembly sequence monitoring method based on multi-view semantic change detection in assembly images is proposed.This method innovatively introduces a densely connected jump fusion mechanism,which enriches the amount of network feature information;studies a self-attention mechanism that integrates contextual feature information,makes full use of contextual feature information to guide dynamic attention matrix learning,and improves the Monitoring performance of network architecture;proposed a semantic step recognition module,which can effectively identify the current assembly stage of changing parts;created three types of mechanical assembly multiview semantic change detection data sets,and based on this experiment.The experimental results show that the assembly sequence monitoring method based on multi-view semantic change detection of assembly images proposed in this paper can effectively improve the detection ability of small target parts,and can be used for mechanical assembly sequence monitoring.(3)A multi-view pose estimation method for mechanical assembly parts based on deep learning is proposed.First,the mechanical assembly pose estimation dataset is built,and then the Dense Fusion-based pose estimation experiment is carried out.Finally,the experimental comparison with the PV Net(Pixel-wise Voting Network)pose estimation network is carried out.The experimental results show that the multi-view pose estimation method of mechanical assembly parts based on deep learning proposed in this study can estimate the pose of assembly parts with low texture and single color from different perspectives,and effectively judge the pose of the parts in the space coordinate system.information,which can be applied to mechanical assembly attitude monitoring.This topic takes mechanical assembly process as the research object,establishes the data set of mechanical assembly multi-perspective change detection and pose estimation,and studies the method of mechanical assembly multi-perspective change detection and pose estimation based on deep learning,which is of certain significance to realize the intelligent monitoring of mechanical product assembly process and promote intelligent manufacturing.