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

作　　者：陈子宜 叶锋[1,2,3] CHEN Ziyi;YE Feng(College of Computer and Cyber Security,Fujian Normal University,Fuzhou 350117,China;Digital Fujian Big Data Security Technology Institute,Fuzhou 350117,China;Fujian Provincial Engineering Research Center of Public Service Big Data Analysis and Application,Fuzhou 350117,China)

机构地区：[1]福建师范大学计算机与网络空间安全学院,福建福州350117 [2]数字福建大数据安全技术研究所,福建福州350117 [3]福建省公共服务大数据挖掘与应用工程技术研究中心,福建福州350117

出　　处：《福建师范大学学报(自然科学版)》2025年第2期35-42,共8页Journal of Fujian Normal University:Natural Science Edition

基　　金：国家自然科学基金面上项目(62072106);福建省创新战略研究计划项目(2023R0156)。

摘　　要：由于城市车载激光点云应用场景的复杂性,导致基于深度学习的语义分割模型面临着目标语义层面的域迁移现象,通常需要重新训练完整模型才能处理新增的语义类别。然而城市车载激光点云通常包括极大量的点,重新训练语义分割模型会浪费大量的资源。通过基于特征记忆库的城市车载激光点云域自适应语义分割方法,解决城市车载激光点云之间的目标语义域迁移现象,使得新增语义类别数据时,只需提取新增语义类别的特征,而无需重新训练完整的语义分割模型,得到的语义分割性能对比重新训练完整模型仅有较小的损失。Due to the complexity of urban vehicle-mounted laser point cloud application scenarios,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 semantic 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 laser 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 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]