摘要
针对三维点云特征提取困难的问题,提出一种以门控网络(GateNet)为基础的PointNet++改进网络,通过抑制无关特征,强调重要特征,自适应校准网络内部特征,达到增强局部特征的目的。将门控网络、挤压激励、注意力机制等引入PointNet++点云分类分割网络,提高分类分割精度,在数据集ModelNet、ShapeNet和S3DIS上进行实验。实验结果表明,改进网络提高了总体分类精度(OA)和均交并比(mIoU),结果优于PointNet++网络,时间更快。
Aiming at the difficulty of feature extraction of 3D point cloud,an improved PointNet++network based on GateNet was proposed to enhance local features by suppressing irrelevant features,emphasizing important features and adaptively calibrating the internal features.The classification and segmentation accuracy of PointNet++is improved by introducing GateNet,SENet and attention mechanism to PointNet++for 3D point cloud.Experiments were carried out on data sets ModelNet,ShapeNet and S3DIS.Results show that the improved network improves the overall classification accuracy(OA)and the mean intersection over union(mIoU),which is better and faster than PointNet++network.
作者
刘慧
田帅华
LIU Hui;TIAN Shuai-hua(School of Electrical and Information Engineering,Beijing University of Civil Engineering and Architecture,Beijing 100044,China)
出处
《计算机工程与设计》
北大核心
2024年第5期1557-1564,共8页
Computer Engineering and Design
基金
国家自然科学基金项目(62170220)。
关键词
深度学习
机器视觉
三维点云
门控网络
挤压激励
注意力机制
分类分割
deep learning
machine vision
3D point cloud
GateNet
squeeze-and-excitation
attention mechanism
classification and segmentation