A Feature Extraction and Combination Method for Tree Detection of Terrestrial Laser Point Clouds in Forest
To address the problems that the methods based on point cloud features for tree detection lack the importance assessment and selection of features,a method of feature extraction and combination of voxel features involved in tree detection is proposed.Firstly,a set of voxel features are proposed based on the structural differences and spatial distribution characteristics of different categories in the forest,and the trunk points are extracted by combining the commonly used basic features such as 3D,2D,grid and intensity.According to the feature importance,the feature combination is constructed and the tree detection is carried out,and then a feature selection method based on the tree detection results is proposed to determine the feature combination suitable for the current sample plot.Experiments were conducted on three different forest sample plots,and the results showed that the trunk point extraction accuracy increased by 7.3％,11.6％,and 9.5％,and single wood detection accuracy increased by 0.4％,2.6％,and 0.7％,respectively,after adding voxel features;compared with the feature selection method based on single point classification results and the feature combinations in other studies,the method in this paper was able to achieve similar or higher tree detection accuracy with fewer features.
NETL
NSTL
万方数据
