Airborne LiDAR Point Cloud Data Classification Based On Improved LBET And Neural Network

In this paper,for the multi-feature fusion of airborne LiDAR point cloud classification,there are errors in sample data and sample selection,resulting in slow classification and low accuracy of classification results.Based on improved LBET(Learning Based on Eigenvalue Transition,LBET)And the airborne LiDAR point cloud classification method of neural networks.The method selects five kinds of classification feature information: ground height,restored reflection intensity,image classification result,return of number and elevation texture.The improved LBET model analyzes the binary signal,obtains the ground standard binary signal and the fuzzy signal,and then uses BP.(Back Propagation,BP)Neural network training standard landmark binary signals and identifying fuzzy signals to implement point cloud classification.Compared with multi-feature fusion classification,the processing time of this method is faster and the precision is obviously improved.The main contributions of this article are as follows:(1)Analysis and processing of point cloud classification features,the reflection intensity is affected by the instrument and the outside world,and the reflection intensity is poorly applied to the point cloud classification.This paper uses the similarity clustering algorithm according to the elevation and reflection intensity.Point cloud reflection intensity repair,in the elevation texture analysis of point cloud,this paper uses the new method to measure the elevation texture.For the acquisition of ground height,the distance from the DTM(Digital Terrain Model,DTM)is proposed as the ground height.(2)An airborne LiDAR point cloud classification method with improved LBET and BP neural network fusion is proposed.Experiments verify the advantages and disadvantages of single neural network and decision tree model classification.In this paper,the two models are optimized and combined.According to the confidence interval of point cloud classification feature information,the classification feature information is segmented and recombined.Based on this,BP neural network is used to complete the point cloud classification.(3)The classification results are evaluated using the Kappa coefficient.By comparing the multi-feature fusion classification results,TerraSolid software classification results and experimental results,the reliability and effectiveness of the classification method are verified.(4)Based on the open source cloud data processing software CloudCompare,this paper implements the proposed algorithm.The main functions of the program are: point cloud import and display,reflection intensity repair,point cloud and image registration fusion,BP neural network classifier.The experimental results show that the proposed algorithm can classify point clouds into tall trees,buildings,low plants,bare ground and roads,and the average accuracy of Kappa coefficient is 87.2%.Compared with other classification methods,the improved LBET and BP neural network fusion algorithm has higher point cloud classification accuracy and has certain applicability.Compared with the multi-feature classification method,the classification speed of the method is faster.