Journal of Computer Applications ›› 2024, Vol. 44 ›› Issue (4): 1086-1092.DOI: 10.11772/j.issn.1001-9081.2023050588

• Artificial intelligence • Previous Articles    

Point cloud semantic segmentation based on attention mechanism and global feature optimization

Pengfei ZHANG1, Litao HAN1,2(), Hengjian FENG1, Hongmei LI1   

  1. 1.College of Geodesy and Geomatics,Shandong University of Science and Technology,Qingdao Shandong 266590,China
    2.Key Laboratory of Geomatics and Digital Technology of Shandong Province,Qingdao Shandong 266590,China
  • Received:2023-05-16 Revised:2023-06-12 Accepted:2023-06-21 Online:2023-08-01 Published:2024-04-10
  • Contact: Litao HAN
  • About author:ZHANG Pengfei, born in 1998, M. S. candidate. His research interests include 3D point cloud data processing, point cloud semantic segmentation.
    HAN Litao, born in 1978, Ph. D., associate professor. His research interests include 3D GIS, indoor positioning and navigation, emergency evacuation planning.
    FENG Hengjian, born in 1998, M. S. candidate. His research interests include video processing.
    LI Hongmei, born in 1998, M. S. candidate. Her research interests include emergency evacuation planning.
  • Supported by:
    National Natural Science Foundation of China(42271436);Shandong Provincial Natural Science Foundation(ZR2021MD030)


张鹏飞1, 韩李涛1,2(), 冯恒健1, 李洪梅1   

  1. 1.山东科技大学 测绘与空间信息学院,山东 青岛 266590
    2.山东省基础地理信息与数字化技术重点实验室,山东 青岛 266590
  • 通讯作者: 韩李涛
  • 作者简介:张鹏飞(1998—),男,河南驻马店人,硕士研究生,主要研究方向:三维点云数据处理、点云语义分割
  • 基金资助:


In the 3D point cloud semantic segmentation algorithm based on deep learning, to enhance the fine-grained ability to extract local features and learn the long-range dependencies between different local neighborhoods, a neural network based on attention mechanism and global feature optimization was proposed. First, a Single-Channel Attention (SCA) module and a Point Attention (PA) module were designed in the form of additive attention. The former strengthened the resolution of local features by adaptively adjusting the features of each point in a single channel, and the latter adjusted the importance of the single-point feature vector to suppress useless features and reduce feature redundancy. Second, a Global Feature Aggregation (GFA) module was added to aggregate local neighborhood features to capture global context information, thereby improving semantic segmentation accuracy. The experimental results show that the proposed network improves the mean Intersection?over?Union (mIoU) by 1.8 percentage points compared with RandLA-Net (Random sampling and an effective Local feature Aggregator Network) on the point cloud dataset S3DIS, and has good segmentation performance and good adaptability.

Key words: 3D point cloud, deep learning, semantic segmentation, attention mechanism, feature aggregation


在基于深度学习的三维点云语义分割算法中,为了加强提取局部特征细粒度能力和学习不同局部邻域之间的长程依赖性,提出一种基于注意力机制和全局特征优化的神经网络。首先,通过加性注意力的形式设计单通道注意力(SCA)模块和点注意力(PA)模块,前者通过自适应调节单通道中各点特征加强对局部特征的分辨能力,后者通过调节单点特征向量之间的重要程度抑制无用特征并减少特征冗余;其次,加入全局特征聚合(GFA)模块,聚合各局部邻域特征,以捕获全局上下文信息,从而提高语义分割精度。实验结果表明,在点云数据集S3DIS上,所提网络的平均交并比(mIoU)相较于RandLA-Net(Random sampling and an effective Local feature Aggregator Network)提升了1.8个百分点,分割性能良好,具有较好的适应性。

关键词: 三维点云, 深度学习, 语义分割, 注意力机制, 特征聚合

CLC Number: