Research On Classification Method Of Airborne Lidar Point Cloud Data

Aiming at the single feature contained in the original point cloud data and the low classification accuracy of traditional classification methods for airborne lidar data,this paper proposes a method of feature extraction based on the point cloud data,and then uses the combined classifier to classify the point cloud.This method mainly extracts the elevation features,intensity features,normal vector features and texture features of the point cloud data.Considering the overall information of the point cloud data,it also combines the point cloud segmentation segment features.On this basis,the combined classifier composed of random forest,BP neural network and support vector machine is used to realize the classification of airborne laser point cloud data.The experimental results show that the classification accuracy of the combined classifier is better than the traditional machine learning method.The main work in this paper includes:(1)After denoising the airborne laser point cloud data,cloth simulation and Triangulated Irregular Network filtering algorithm are applied to filter the point cloud data,and the results are analyzed to select the most suitable filtering method.(2)Feature extraction of point cloud data;It mainly includes the extraction of neighborhood feature,normal vector feature,texture feature and segment feature.Neighborhood feature is the feature obtained by searching the neighborhood of point cloud data and using scientific theoretical methods;The feature of normal vector of point cloud is based on local surface fitting algorithm.Firstly,plane fitting is performed on the point cloud data,and then the normal vector is obtained;Segmentation segment feature is mainly based on the data of point cloud after denoising.The whole data is divided into multiple point cloud sets with similar attributes.On this basis,the segmentation segment feature is obtained by calculating the feature;The texture features of point cloud are mainly obtained by using scientific theory to obtain the texture features related to the elevation and intensity information of point cloud data.(3)Point cloud classification;The extracted point cloud features are integrated as the input data of point cloud classification,and the combined classifier composed of random forest,support vector machine and BP neural network is used to classify the point cloud data,and the accuracy of the model is evaluated.The experimental results show that the airborne laser point cloud data can be classified into six classifications,mainly including high vegetation,low vegetation,buildings,roads,grassland and other classifications,based on the point cloud feature extraction and combined classifier classification method.The accuracy evaluation index kappa coefficient calculated by confusion matrix is 91.76%,Compared with single machine learning classification algorithm,the classification accuracy of combiner classifier is improved.