Semi Supervised Point Cloud Classification Algorithm Based On Multi-scale And Multi-level Features

LiDAR(light detection and ranging)is an effective way to copy the urban scene data through aerial photogrammetry remote sensing.It plays an important role in the real 3D city construction and planning,location navigation and virtual reality service.The processing and deep processing of LiDAR point cloud data is the precondition of engineering application and scientific research,which has become an important research direction in the field of Geographic Information Science in recent years.The difficulties of lidar data classification and processing lie in: on the one hand,the LiDAR point cloud data of urban scene contains rich and complex data features,different shapes,complex structures,mutual occlusion,resulting in large data loss;on the other hand,the airborne lidar scanner is far away from the characteristics of urban targets,such as cars,pedestrians,etc.,in motion during the scanning process,resulting in the acquisition The density of lidar points taken is uneven and there are loopholes.The purpose of this paper is to construct a feature parameter with high and stable differentiation to the surface objects by fusing the multi-scale and multi-level point set features of the laser point cloud data,and to study the point cloud data classification algorithm using artificial intelligence.In this algorithm,the point cloud data are divided into four categories: ground points,vegetation,buildings and cars.First of all,the acquired data are spliced,system error corrected and denoised to remove the factors that affect the post-processing of point cloud data,and then the pre-processing data are filtered to effectively separate the ground point and non-ground point.Then,the non surface point data set is used as the research object,and the multi-scale point set features are constructed based on the difference of point cloud data characteristics and spatial scope of the ground object entities.The point cloud is aggregated into multi-level point set based on the threshold of natural index function,and the shape features of each point set in the multi-level point set are extracted by combining sparse coding and potential Dirichlet distribution.Finally,the appropriate selection is made In this paper,we use the three-dimensional spatial information data of different scene environment to test the classification algorithm.The experimental results show that the algorithm has high classification accuracy,fast processing speed and small noise impact,which can effectively solve the problem of complex city scene point cloud data classification,and has important significance in the field of artificial intelligence lidar data classification.