地理空间信息2025，Vol.23Issue(4)：20-24,5.DOI:10.3969/j.issn.1672-4623.2025.04.005
顾及回转窑几何结构特征的点云语义分割
Semantic Segmentation of Point Clouds Considering the Geometric Structural Characteristics of the Rotary Kiln
郑琦 1邹进贵 1贺亦峰 1翟若明 1程志新1
作者信息
- 1. 武汉大学测绘学院,湖北武汉 430079
- 折叠
摘要
Abstract
Aiming at the problems of the traditional point cloud semantic segmentation deep learning method in the rotary kiln wheel belt recogni-tion,such as the small number of training samples,the lack of geometric features,and the sparse proportion of wheel belt structure,we proposed a rotary kiln point cloud segmentation method based on data augmentation and multi-feature description.We used it as the original data to con-struct the rotary kiln point cloud dataset,and used PointNet,PointNet++and PointNetCNN network to train the segmentation model for wheel belt extraction of the rotary kiln.We improved the local feature expression ability of point cloud by designing the curvature and normal vector geometric feature descriptors,and used the data augmentation method to solve the problem of small number of samples and the unbalanced num-ber of semantic class samples.In order to verify the proposed method,we compared and analyzed the influence of designed geometric descriptors and data augmentation methods on the results of semantic segmentation deep learning in PointNet,PointNet++and PointCNN network models.The three accuracy indexes of OA,mIoU and mAcc show that the segmentation effect of cylinder,wheel belt and miscellaneous points in the rota-ry kiln has been improved,which provides a practical and effective solution for subsequent rotary kiln related projects.关键词
回转窑/激光点云/数据增强/多特征描述/点云语义分割方法Key words
rotary kiln/laser point cloud/data augmentation/multi-feature description/semantic segmentation method of point cloud分类
天文与地球科学引用本文复制引用
郑琦,邹进贵,贺亦峰,翟若明,程志新..顾及回转窑几何结构特征的点云语义分割[J].地理空间信息,2025,23(4):20-24,5.
