Individual Tree Species Classification Using Airborne Hyperspectral Imagery And Lidar Data

Abstract

In this study high spatial resolution (1 m) hyperspectral images and LiDAR data (8p/m ² ) was applied to discriminate among tree species of mixed forest. The main objective of this study was to apply machine learning methods using crown segments for image classification. A watershed segmentation algorithm was used to delineate individual crowns from a filtered CHM model. The image classification was applied on the original spectral bands and transformed (MNF) dataset. A binary tree SVM classifier was developed in accordance with the principle of SVM, based on the Jeffries-Matusita (JM) separability measure of selected classes. The ABTSVM on MNF-transformed dataset provided more accurate results than applied multiclass SVM methods. The addition of crown segments resulted in an increase in classification accuracy of 14.51 percentage points over pixel-based classification alone.