Point cloud semantic segmentation is the basic step of 3D point cloud data processing,and is the key link of 3D scene understanding analysis,reconstruction and target recognition.In view of the current problem of lack point cloud information and low accuracy used for semantic segmentation of 3D point clouds,this paper adds point cloud RGB information and normalized coordinate information of the room to which it belongs based on the use of point cloud 3D coordinate information to enrich the amount of information at the input end of the neural network and further improve the segmentation accuracy of the model.Finally,PointNet++ is used to test the improved 3D point cloud semantic segmentation effect.The experiment shows that after enriching the data information and inputting the network,the model accuracy is improved by 6.65 %.