关键词: 点云, 分类分割, 自适应, 动态图卷积, 注意力机制

Abstract: To address the challenges of traditional convolution in accurately extracting neighborhood feature information and effectively integrating contextual information in point cloud processing, a Point Cloud Classification and Segmentation Method Based on Adaptive Dynamic Graph Convolution and Parameter-Free Attention was proposed. Firstly, the Adaptive Dynamic Graph Convolution module was used to learn the feature information of different neighborhoods, generate the convolution kernels, and update the edge features, thereby to accurately extract the local neighborhood features of the point cloud. Then, a residual structure was designed to learn the spatial position information of the point cloud, so as to accurately capture the geometric structures between the point pairs, and better retain and extract the detailed features. Finally, in order to better pay attention to and extract local geometric features, the Parameter-Free Attention (PFA) module was combined with convolution operation to enhance the connection between neighbors and improved the context-aware ability of the model. Experimental results show that, compared to the PointNet, the proposed method has significant significant advantages across various tasks: an increase of 4.6 percentage points in Overall Accuracy (OA) for classification tasks, a 2.3 percentage point increase in mean Intersection over Union (mIoU) for part segmentation tasks, and a 24.6 percentage point increase in mIoU for semantic segmentation tasks. The proposed method further improves the understanding and characterization of complex geometries, resulting in more accurate feature extraction and experimental performance in a variety of tasks.

Key words: Keywords: point cloud, classification and segmentation, adaptive, dynamic graph convolution, attention mechanism