Research On Deep Learning Point Cloud Classification Based On Fused Graph Convolutio

Because of its strong spatial expression ability,point cloud data has become an important data source for surveying and mapping geographic information,computer vision,driverless and other industries.With the maturity of sensor technology and the popularity of various 3D laser scanners,the collection of point cloud data tends to be convenient and popular,which promote the use of point cloud data.Point cloud classification is the basis of point cloud data application,which has become not only an important research direction,but also a research hotspot and difficulty,the irregular structure of point cloud makes the high-precision classification of point cloud very challenging.Traditional point cloud classification methods have poor semantic feature expression ability and robustness.With the development of deep learning,the feature extraction of point cloud classification has changed from manual design extraction and point cloud projection extraction to point cloud data extraction.In view of the defects of the existing point cloud classification methods,such as being insensitive to the local area of the point cloud and lack of effective feature description,a point cloud classification network based on fusion graph convolution is studied,the details are as follows:(1)By constructing the graph structure of spatial domain and spectral domain to extract two local features of point cloud,and the sensitivity of the algorithm to the local area of point cloud is improved by feature fusion,so as to improve the problem of lack of local information description in the process of point cloud feature extraction.(2)Propose a deep learning point cloud classification network based on fusion graph convolution,introduce spatial pyramid pooling to deepen the fine-grained description of fusion features and obtain more robust feature representation,improve the performance of classification network.(3)The classification experiments are carried out on the lidar scanning point cloud data of outdoor scene and the point cloud data of indoor scene respectively,good classification results are obtained in both scenarios,which indicates that the fusion graph convolution network has good generalization.Compared with the existing methods,the experimental results highlight the effectiveness and advantages of the fusion graph convolution method.