Tree-level Structure Characterization Based On Static Terrestrial Laser Scanning Point Clouds

Li DAR, as a kind of active remote sensing technology, can get high precision of three-dimensional spatial information especially in forest height and vertical structure. Accurate tree-level characteristic information is increasingly demanded for forest management and environment protection. The cutting-edge remote sensing technique of terrestrial laser scanning(TLS) shows the potential of filling this gap.This study first focuses on exploring the methods for deriving various tree stem structural parameters, such as stem position, diameter at breast height(DBH), tree height, the crown diameter, the degree of stem shrinkage, and the elevation angle and azimuth angle of stem inclination. Second, converting the stem point cloud to the plane, the texture parameter is calculated by the first and second order, analyzing the mean and standard variance in the first statistical analysis and using gray level co-occurrence matrix for energy, contrast, correlation and homogeneity calculation, respectively.In the end, the study is exploring a method for tree species classification based on the tree structure parameters and texture parameters. Only with a single tree structure parameters cannot separate tree, but preliminary conclusion is that stem shrinkage can be used as a classification factor. Therefore, the support vector machine(SVM) classification is introduced, choosing three times polynomial kernel function of SVM classifier. When one of the factors is the degree of stem shrinkage in two factor analysis, species classification accuracy is lower; however, over two factors in tree species classification the accuracy is higher than that of two factors. Based on the parameters in the first-order statistical analysis, the average accuracy is only 58%. With the SVM classifier and the four characteristic parameters in gray level co-occurrence matrix, the final accuracy has been improved, up to 70%. Limited to the study area and reference data, although accuracy is better, further study of tree species classification is necessary only in combination with the TLS point cloud.In summary, the results show that the method of derivation of tree structure and texture parameters is preliminarily, which is helpful for driving application of TLS in the forest resource investigation.