UAV-Based Hyperspectral Imaging System for Tree Species Identification in Tropical Forest of Malaysia

Abstract

Hyperspectral data is usually used as the main source of data for remotely sensed tree species identification. However, the number of hyperspectral sensors available in Malaysia is limited. This study aims at evaluating the performance of unmanned aerial vehicle (UAV)-based OCITM-F hyperspectral imager sensor in identifying six tree species on the campus of Forest Research Institute Malaysia (FRIM). This study attempts to compare data types and classification approaches to find the best way to identify the tree species. Besides reflectance (R) and derivative (D) spectra, a log of spectral reflectance (LogR) and derivative of the log of spectral reflectance (DLogR) were evaluated using Random Forest (RF) and Support Vector Machine (SVM) classifiers. Boruta technique was also used to reduce the dimensionality of the data input. The results demonstrate that the performances varied with different combinations of data input and classifier. Reflectance spectra classified with SVM classifier gave the highest accuracy of 72.6% (Kappa 0.6421), while the lowest accuracy was achieved with the combination of derivative spectra and RF classifier with an accuracy of 52.9% (Kappa 0.4026). Based on this study, the UAV-based OCITM-F hyperspectral imager sensor has the potential to be used to identify forest tree species in a tropical forest with good accuracy.