Classification of Tree Species Using Point Cloud Features from Terrestrial Laser Scanning

Abstract

The automatic classification of tree species using terrestrial laser scanning (TLS) point clouds is key in forestry research. This study aims to develop a robust framework for tree species classification by integrating advanced feature extraction and machine learning techniques. Such a framework is of great significance for investigating and monitoring forest resources, sustainable forest management, and biodiversity research. To achieve this, point cloud data from 360 trees of four species were collected at the Northeastern Forestry University in Harbin City, Heilongjiang Province. Three types of tree point cloud features were extracted: tree structure, bark texture, and bark color. In addition, to repair and optimize the bark point cloud data, improved bark texture features were generated using the kriging interpolation method. These four features were combined into seven classification schemes and input into a random forest classifier, which was used to accurately classify the tree species. The results showed that the classification scheme combining tree structure features, improved bark texture features, and bark color features performed the best, with an overall classification accuracy of 94.17% and a kappa coefficient of 0.92. This study highlights the effectiveness of integrating point cloud data with machine learning algorithms for tree species classification and proposes a feature extraction and classification framework that significantly enhances classification accuracy.