In order to improve the classification accuracy of 3D point cloud data,the fusion algorithm of PCT with attention mechanism and PointMLP with residual point MLP is investigated.The improved algo-rithm firstly adds a geometric affine module to normalize the input point cloud to solve the density inhomo-geneity and geometrical uncertainty of the point cloud;then the coding module firstly uses the attention mechanism to globally express the upper layer features,and then adds a residual point MLP module to re-present the depth features of the output features of the attention module to fully extract the point cloud fea-tures;and finally,the classification module to complete the downstream point cloud classification task.The experimental results show that the improved algorithm achieves a classification accuracy of 95.6%on the ModelNet40 dataset,which is a 2.4%increase in accuracy compared to the PCT model,and is robust to changes in the number of point clouds.
3D point cloudattention mechanismresidual point MLPpoint cloud classification
维普
万方数据
