摘要：This article proposes a point cloud classification model based on group normalization to increase the classification accuracy when the computing power of the terminal device is limited. This model groups and normalizes the features of point cloud during inference and increases the classification accuracy when the computing power is limited. The group normalization first groups the features of point cloud by their channel, then computes their statistic metrics and normalizes them. Also, one-dimensional convolution layers are used to replace the fully connected layers to decrease the model parameters and keep the model's performance when the computing power is limited. In the experiment, PointNet++ is used to pretrain on ModelNet40 and then fine-tune on the point cloud data of transmission lines. The result shows that the proposed method can effectively increase the classification accuracy and help the 3D modeling process of the transmission line.
