Three-dimensional point cloud classification method based on graph convolution and shape descriptor

Abstract

The invention provides a three-dimensional point cloud classification method based on graph convolution and a shape descriptor. The shape descriptor is added into a graph convolution neural network. According to the method, firstly, the relation between points in a point cloud is represented by an adjacent matrix according to a general flow of graph convolution, and then the features of neighborhood spaces of the points are described by using shape descriptors and are added into the graph convolution as the local features of the three-dimensional point cloud. Defects of spectral domain graph convolution on local feature processing are made up to a certain extent. Then original features and shape descriptors are combined together, a graph convolutional neural network is used for aggregation, and finally a final classification result is obtained. The shape descriptor and the graph convolution are combined together, so that the graph convolution operation can effectively aggregate local features, more comprehensive information representation is obtained, and the classification result of the three-dimensional point cloud can be better improved.

Description

Technical field: [0001] The invention relates to a classification method based on graph convolution and three-dimensional shape descriptors, and the method has good application in the field of point cloud processing or three-dimensional model processing. Background technique: [0002] At present, with the continuous development of deep learning, the traditional convolutional neural network used to process Euclidean space data has fallen into a bottleneck in some fields, and the graph convolutional neural network used to process non-Euclidean space data has gradually emerged. And it has a good momentum of development, and has a very wide range of applications in various fields such as traffic prediction, natural language processing, computer vision and other fields. In the past, for the disorder and rotation of point clouds, the PointNet series of neural networks were often used to perform point cloud classification and segmentation tasks, and the use of graph convolutional n...

AI Technical Summary

Problems solved by technology

In the past, for the disorder and rotation of point clouds, the PointNet series of neural networks were often used to perform point cloud classification and segmentation tasks, and the use of graph convolutional neural networks can also be well avoided due to the disorder and rotation of point clouds. However, compared with the PointNet series of neural networks, the graph convolutional neural network is essentially continuously aggregating node information to obtain the final global features, which leads to a certain graph convolutional neural network. To a certain extent, the extraction of local features is ignored, so there is still room for improvement in the method of simply using graph convolution to classify point clouds

Images