Laser point cloud classification is the basis for 3D scene understanding. In order to solve the problems of insufficient feature expression and unbalanced sample categories in the classification of large scenes of airborne point clouds， this paper proposed an airborne point cloud classification method that integrates edge convolution and global-local self-attention. Firstly， U-net was used as the network framework to integrate Point Transformer and edge convolution modules， so that the model could pay attention to complex ground object boundaries and texture information， and obtain local geometric features with better expressive ability. Secondly， a self-attention mechanism that integrates global context information and local structural features was innovatively proposed， and the global self-attention module tended to the information of the entire input sequence， while the local self-attention module pays more attention to the details of the local region. The combination of the two mechanisms enhanced the capture of long-distance dependence and local structure， and at the same time， the model could take into account the key features of a few categories， reduced the impact of sample category imbalance on the classification accuracy to a certain extent， and helped to improve the classification ability of the model for complex ground object relationships. Finally， the proposed method was verified on the public ISPRS-3D dataset and WHU-Urban3D dataset， and the experimental results show that the classification accuracy of the proposed method on the two datasets is 82.5% and 87.4%， respectively， which is better than that of the classical networks such as PointNet++ and Stratified Transformer and the ISPRS 3D official website competition network， which can effectively improve the classification accuracy of airborne point clouds.