Design of neural network model for lightweight 3D point cloud classification

In this paper, a lightweight point cloud classification model, LightSeDgCNN, is proposed to solve the problem of large number of basic parameters and long training time. Based on the latest DGCNN model, the structure of the model is optimized through experiments. Secondly, the influence of model classification power reduction caused by the reduction of lifting parameters is improved by introducing the SENeT model (which can automatically acquire the importance of each feature channel through learning). Finally, the experiment is carried out on the ModelNet40 data set. The experimental results show that LightSeDgCNN can achieve the same effect of point cloud classification and recognition with less model parameters and training time.