Abstract:Accurately detecting the changed parts of the mechanical assembly during the assembly process is of great significance for monitoring the assembly sequence of the product,improving the assembly quality and ensuring production safety.In order to detect the changing parts of mechanical assembly from multiple angles,a multi-view change detection method (TAF Net) of mechanical assembly images based on three-dimensional attention and bilateral filtering network was proposed.In order to improve the accuracy of TAF Net detection of mechanical assembly changes,a 3D attention mechanism was introduced to enhance the networks detailed feature extraction capability;bilateral filtering was introduced to reduce noise in the images and optimize the boundaries of parts in the changed images.Two assembly change detection data sets were established,which were synthetic depth image data set and real color image data set.The two data sets were used for experiments.The experimental results show that the TAF Net network can accurately detect the change area in the images,the comprehensive evaluation index F1_s core in the two data sets all reach more than 96%.