Abstract: A part model feature recognition method combining attribute adjacency graph and point cloud was proposed by combining two feature recognition methods to overcome the limitations of current part model feature recognition technology based on attribute adjacency graph and point cloud.The model attribute adjacency graph was used to match feature subgraphs to find and separate feature surfaces,and then the feature surfaces in point clouds were sampled.The point cloud classification network structure on the basis of PointNet network was improved by adding a local feature extraction module and a Transformer based non-local feature extraction module and combining feature attribute adjacency graph information with original point cloud data.Experimental results indicate that the recognition accuracy for 24 common features is 99.92 %.

Key words: part model feature recognition, attribute adjacency graph, point cloud, Transformer net, PointNet