| 摘要: |
| 点云分割是场景理解、目标识别、文化遗产保护等领域的一项基础且关键的技术。然而，由于点云数据的复杂性，如何从点云中提取深层特征对目前的研究提出了很大的挑战。为了解决这一问题，提出了一个结合超图卷积和生成对抗网络的新框架，用于点云分割。首先，使用超图建模点云之间的几何拓扑关系，捕捉点云中的高阶相关性。其次，以多层超图卷积作为鉴别器，结合生成对抗模型构建超图生成对抗网络，提取点云稀疏区域的细节特征。最后，利用逐点损失和对抗损失相结合对网络进行训练，提升网络训练的稳定性和标签预测的准确性。在ShapeNet Part数据集上进行了点云分割实验，结果表明所提方法在16个类别上的平均交并比（Intersection over Union，IoU）为84.5%，较大地提升了分割精度。 |
| 关键词: 三维点云分割 超图卷积神经网络 生成对抗网络 |
| DOI：10.20079/j.issn.1001-893x.250127001 |
|
| 基金项目:广西科技计划项目(桂科AD23026225)资助；广西自然科学基金青年科学基金项目(2025GXNSFBA069227)；认知无线电与信息处理教育部重点实验室基金项目(CRKL230205) |
|
| Hypergraph Convolutional Generative Adversarial Network for 3D Point Cloud Segmentation |
| TIAN Fangyuan,CHEN Feng,ZHU Guifa,NONG Liping |
| (1.School of Physics and Technology，Guangxi Normal University，Guilin 541004，China; 2.Key Laboratory of Cognitive Radio and Information Processing，Ministry of Education，Guilin 541004，China; 3.School of Information and Communication，Guilin University of Electronic Technology，Guilin 541004，China) |
| Abstract: |
| Point cloud segmentation is a fundamental and key technology in scene understanding，object recognition，and cultural heritage protection.Due to the complexity of point cloud data，how to extract deep features from point clouds poses a great challenge for the current research.To address this challenge，a novel framework combining hypergraph convolution and generative adversarial network is proposed for point cloud segmentation.Firstly，higher-order correlations in the data are captured by modeling the geometric topological relationships between point clouds using a hypergraph.Secondly，multiple hypergraph convolutional layers are used to build a discriminator，and combined with the generative adversarial model，a hypergraph generative adversarial network is formed to extract the detailed features of the sparse areas of point clouds.Finally，the network is trained by a combination of point-wise loss and adversarial loss，which can improve the stability of the network training and the accuracy of label prediction.Point cloud segmentation experiments are conducted on the ShapeNet Part dataset and the results show that the proposed method achieves an average intersection over union(IoU) of 84.5% on 16 categories，which greatly improves the segmentation accuracy. |
| Key words: 3D point cloud segmentation hypergraph convolutional neural network generative adversarial network |
《电讯技术》编辑部 