Abstract: To enhance the application effect of point cloud technology,an efficient point cloud classification method is proposed. Specifically,the point cloud dataset is preprocessed,spatial domain graph convolution is used to extract feature information of local regions,then spectral domain graph convolution is employed to extract global feature information,and finally,point cloud classification is completed through the fully connected layer. The experimental results show that the point cloud classification technology that combines spatial domain graph convolution and spectral domain graph convolution performs best in image segmentation tasks,with a mean intersection over union （mIoU） of 49.8 at convergence. In the point cloud classification test,the research model achieves an accuracy of 97.2% in the classification of low vegetation,which is superior to similar models. It can be seen that this technology has a good application effect and is conducive to promoting better deployment and application of point cloud technology.

Key words: point cloud classification, graph convolution, features, preprocessing