Point cloud segmentation method of satellites based on improved DGCNN

Using the satellite characterization information obtained by the space-based platform, the key parts of the satellite, such as solar panels, satellite payload, and propulsion systems, are segmented. The target object segmented from the point cloud data is significant to improve the accuracy of subsequent point cloud registration and attitude recognition. In this study, we introduced TSNet, which has the following characteristics. 1) The continuous recursive gate convolution module (g_nConv) is introduced into the network, which can improve the accuracy of point cloud segmentation. 2) The weight channel for feature transfer is designed to avoid global information loss. The mIoU value of TSNet laser point cloud segmentation reached 88.12%, which was better than common point cloud segmentation algorithms, such as PointNet, PointNet++ and DGCNN. The proposed method can provide more accurate perception information for ground control personnel.