Lidar point cloud classification method based on improved PointNet++

LiDAR point cloud classification is a key step in extracting valuable information from massive point cloud data. In order to improve the accuracy of point cloud classification based on deep learning methods, on the basis of PointNet++ network, the CBAM attention mechanism is integrated to improve the model’s ability to extract local feature information of point clouds. The extracted features are processed by the channel and spatial attention module, so that the model can better focus on global context information and effectively suppress noise interference. Secondly, the residual multi-layer perceptron is introduced to reduce the parameters of the network model, which improves the classification accuracy and reduces the training overhead. Finally, training and testing are carried out on the public dataset ModelNet40, and compared with the previously common network models, the overall classification accuracy (OA) and average classification accuracy (mAcc) are improved, and the performance of point cloud classification is effectively improved.