Semantic segmentation of airborne LiDAR point cloud based on global geometric attention network

Deep learning-based semantic segmentation of airborne laser scanning (ALS) point clouds receives extensive attention from researchers in the field of remote sensing and computing. However, due to the special properties of ALS point cloud data, it makes difficulties in the semantic segmentation task. Therefore, a new global geometric attention network (GGAN) for ALS point cloud data semantic segmentation is proposed in this paper. Firstly, a local geometric attention convolution module is proposed to embed local geometric feature information. Then, a global relationship aware attention module was inserted into the network to improve the global perception of the geometric information. The experimental results of semantic segmentation show that the GGPN method can achieve satisfactory classification results. Semantic segmentation experiments are conducted on the ISPRS benchmark point cloud dataset to evaluate the performance of the GGAN network, and the method in this paper achieves an overall accuracy of 81.6 and the average F1 of 71.4 which is better than most of the existing popular methods.