Abstract: To address the problem that existing point cloud segmentation methods ignore directional information when processing local features and cannot effectively extract point cloud neighborhood features due to the limitation of convolutional kernel size, this paper proposed a method that combined both directional encoding and hole sampling to maximize the local receptive field of the network; used the graphical convolutional neural network to exploit the information of the points in the local neighborhood; applied the neighborhood feature extraction layer to automatically weight and fuse the neighborhood features into more representative individual features; combined with the spatial attention mechanism to increase the connection between remote points. The results of object segmentation experiments on the S3DIS dataset show that the OA and mIoU of the proposed method are, respectively, 1.3% and 4.0% higher than PointWeb, 0.6% and 0.7% higher than the baseline method RandLA-Net, and the use of hole sampling and directional coding can effectively improve the semantic segmentation accuracy of point clouds.