A study of point cloud classification methods based on residual attention and gated propagation structure

Point cloud semantic segmentation is a key technology to realize accurate perception and intelligent understanding of 3D environment, which is of great significance to promote the intelligent development in the fields of automatic driving, robot navigation, and industrial inspection. In this paper, a deep learning point cloud classification network is designed to increase the attention to important features by using residual attention mechanism to integrate global features; gated propagation is used between the encoding and decoding layers to reduce the propagation of redundant information in the process of local feature and global feature aggregation, and to improve the expressive power of network features. In order to verify the effectiveness of the method proposed in this paper, this paper takes the point cloud data of underground roadway as an example for experimental verification, and uses the ablation test to verify the effectiveness of the residual attention mechanism. The results show that the method can realize high-precision classification of point clouds, and the residual attention mechanism can utilize the deep global features of the network to effectively improve the classification accuracy of the network for underground roadway point clouds.