AUTHOR=Yin Zhimin , Ji Shichao , Zhang Xuyong , Dai Jianhua , Yu Weiyong , Wu Song 

TITLE=Classification Model of Point Cloud Along Transmission Line Based on Group Normalization

JOURNAL=Frontiers in Energy Research

VOLUME=Volume 10 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/energy-research/articles/10.3389/fenrg.2022.839273

DOI=10.3389/fenrg.2022.839273

ISSN=2296-598X

ABSTRACT=This paper 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 normalize the features of point cloud during inference and increases the classification accuracy when the computing power is limited. The group normalization first group the features of point cloud by their channel, then compute their statistic metrics and normalize them. Also, 1D convolution layers are used to replace the fully connected layers to decrease the model parameters and keep its performance when the computing power is limited. In the experiment, PointNet++ is used to pretrain on ModelNet40 and then finetune on the point cloud data of transmission lines. The result shows that the proposed method can effectively increase the classification accuracy and can help the 3D modeling process of the transmission line.