福建师范大学学报（自然科学版）2025，Vol.41Issue(2)：35-42,8.DOI:10.12046/j.issn.1000-5277.2024060039
基于特征记忆库的三维点云域自适应语义分割
Domain Adaptive Semantic Segmentation for 3D Point Clouds Based on Feature Memory Bank
摘要
Abstract
Due to the complexity of urban vehicle-mounted laser point cloud application sce-narios,deep learning-based semantic segmentation models often encounter domain shift at the target semantic level,typically requiring retraining of the entire model to accommodate newly added se-mantic categories.However,urban vehicle-mounted laser point clouds usually contain an enormous number of points,making full model retraining highly resource-intensive.This article proposes a feature memory-based domain-adaptive semantic segmentation method for urban vehicle-mounted la-ser point clouds to address the target semantic domain shift between urban vehicle-mounted laser point clouds.When incorporating new semantic category data,our approach requires extracting only the features of the new semantic category,rather than retraining the entire semantic segmentation model.The proposed method achieves comparable semantic segmentation performance only a slight loss compared to full model retraining.关键词
三维点云/车载激光/语义分割/域自适应/增量学习/计算机视觉Key words
3D point cloud/vehicle mounted laser/semantic segmentation/domain adapta-tion/incremental learning/computer vision分类
计算机与自动化引用本文复制引用
陈子宜,叶锋..基于特征记忆库的三维点云域自适应语义分割[J].福建师范大学学报（自然科学版）,2025,41(2):35-42,8.基金项目
国家自然科学基金面上项目(62072106) （62072106）
福建省创新战略研究计划项目(2023R0156) （2023R0156）
