河南理工大学学报（自然科学版）2026，Vol.45Issue(5)：87-96,10.DOI:10.16186/j.cnki.1673-9787.2025120086
基于体素双边注意力与八分体编码的点云实例分割
Point cloud instance segmentation based on voxel-based bilateral attention and octree feature encoding
摘要
Abstract
Objectives The essence of point cloud instance segmentation is to generate a closed classifica-tion surface in 3D space through a nonlinear mapping model,so that the target point cloud can be distin-guished.However,the elongated and protruding point clouds of instance objects are usually distributed as outliers,which poses a challenge to accurate point cloud segmentation.Therefore,capturing the correlation features between outlier sparse point clouds and target objects in candidate instance regions,generating pre-cise instance masks and thereby establishing a nonlinear mapping model from positional space to classifica-tion surfaces are considered critical for accurate point cloud instance segmentation.Methods In this paper,a miniature U-Net based on the voxel-based bilateral attention mechanism was constructed.Channel and po-sitional attention mechanisms of point clouds were fused at the voxel level,enabling the model to focus on local positional features while adaptively readjusting feature maps by learning the importance of each chan-nel.In this way,long-range dependencies among point clouds within candidate instances were established.Meanwhile,to address the problem of outlier sparse point cloud misclassification,an octant feature encod-ing module was proposed.Candidate instances were split and encoded into octants,and both the internal characteristics of each octant and the correlation features between octants were extracted.As a result,dis-tant outlier scattered points were incorporated into the feature extraction of candidate instances.Results Ex-perimental results show that the proposed method achieves 51.8%mAP and 68.2%mAP50,demonstrating a significant improvement in point cloud instance segmentation performance.Conclusions More accurate in-stance masks were generated by the bilateral attention mechanism and octant feature encoding module,en-abling precise classification of outlier point clouds.关键词
点云实例分割/双边注意力/八分体特征编码/通道注意力Key words
point cloud instance segmentation/bilateral attention/octant feature encoding/channel attention分类
信息技术与安全科学引用本文复制引用
毕晓华,牛春博,杨艺,王田..基于体素双边注意力与八分体编码的点云实例分割[J].河南理工大学学报（自然科学版）,2026,45(5):87-96,10.基金项目
国家自然科学基金资助项目(92467108) （92467108）
