A single tree extraction method based on spectral clustering algorithm for lidar point cloud data

Abstract

The object of the present invention is to provide a method for extracting a single tree of LiDAR point cloud data based on spectral clustering algorithm. The specific method is: normalize the height information of LiDAR point cloud data, and use mean shift clustering algorithm to perform Construct a similarity graph in voxel space based on the Gaussian similarity function; use the method to calculate the eigenvalues ​​and eigenvectors of the similarity graph, and use the eigenvalue interval to determine the number k of split trees; take the features corresponding to the first k smallest eigenvalues The vector is used to construct the eigenvector matrix as a column, and k-means clustering is performed on the normalized row elements of the eigenvector matrix in the feature space, and the segmentation result is mapped back to the LiDAR point cloud to obtain single-tree clustering, so as to realize the single-tree clustering of the point cloud segmentation. The method provided by the invention can not only perform effective single tree segmentation at the sample plot scale, but also provide stable segmentation results at the regional scale, and has strong practical value.

Description

technical field [0001] The invention relates to a laser radar point cloud data processing technology, in particular to a method for single tree extraction of LiDAR point cloud data based on a spectral clustering algorithm. Background technique [0002] LiDAR (Light Detecting and Ranging, LiDAR) technology is one of the most revolutionary achievements in the field of remote sensing in the past 20 years. As an active remote sensing technology, airborne lidar can obtain the spatial structure characteristics of forests in a large range, and has outstanding advantages in high-precision extraction of key forest parameters. [0003] At present, there are many algorithms for extracting single trees from airborne LiDAR point cloud data, which can be mainly divided into surface model methods based on Canopy Height Models (CHM) and 3D methods using point cloud information. 3D methods can further Divided into clustering methods and voxel-based methods. The clustering method is an effe...

AI Technical Summary

Problems solved by technology

However, these algorithms need to make assumptions about the number of individual trees, and at the same time have certain requirements for the distribution of data, and the segmentation accuracy cannot be guaranteed

Images