基于多尺度动态图卷积网络的3D点云分类

3D POINT CLOUDS CLASSIFICATION BASED ON MULTI-SCALE DYNAMIC GRAPH CONVOLUTION NETWORK

3D点云数据是一种不规则性数据,传统卷积神经网络无法直接对3D点云数据进行处理。对此,提出一种基于多尺度动态图卷积网络的3D点云分类模型。利用最远点采样方法采样3D点云数据集的代表点,降低模型计算复杂度;利用不同尺度的k最邻近节点聚合方式,对图中每一个中心节点的k最邻近节点进行定位;利用边卷积操作对中心节点及其邻接节点的局部属性特征进行提取与聚合用于分类。实验表明,该模型在3D点云分类准确度上,达到了比当前主流模型更高的水平,并且大幅降低了模型生成参数的数量。

Due to the irregularity of 3D point clouds,traditional convolutional neural networks cannot be applied on 3D point clouds processing directly.In view of this,this paper proposes a multi-scale dynamic GCN model for 3D point clouds classification.A farthest point sampling method was applied in our model to efficiently cover the entire point set and decrease model computation complexity.The different scale k-NN group method was used to location k nearest neighborhood for each central node.Edge convolution(EdgeConv)operation was used to extract and aggregate local features between neighbor connected nodes and the central node.The experiments show that our model achieves a better performance on classification accuracy than other state-of-the-art models,and decreases model parameters size dramatically.