LAUSR

Abstract Feature classification has always been one of the research hotspots in scientific visualization. However, conventional interactive feature classification methods rely on prior knowledge and typically require trial and error, whereas feature classification based on data mining is generally based on local features; therefore, obtaining good results with traditional methods is difficult. In this paper, we first map a volume to the super-voxel graph using a 3D extension of the simple linear iterative clustering algorithm and then construct a graph convolutional neural network to implement node classification in a semi-supervised way, i.e., a small number of user-labeled super-voxels. We transform the feature classification of a volume into the classification task of nodes of a super-voxel graph, which is a novel approach and broadens the application scope of graph neural network to volumes. Experiments on different volumes have demonstrated the strong learning ability and reasoning ability of the proposed method. Graphic abstract