Juvenile tree classification based on hyperspectral image acquired from an unmanned aerial vehicle

Abstract

ABSTRACTThe technique of unmanned aerial vehicle (UAV) offers great advantages in vegetation classification because of its high spatiotemporal resolutions and low cost and risk in field measurements. This study explores the potential of using hyperspectral images acquired from UAV to classify juvenile trees, and the feasibility of applying different spectral parameters to classify vegetation. In experimental wooded plots in Hebei province (38.703644°N, 115.400848°E), China, hyperspectral images were obtained from UAV in high spatial and spectral resolutions. The images were corrected for radiative and geometric distortions, and noise was removed. Ten principal components (PCs) and 32 hyperspectral vegetation indices (HVIs) were extracted. Different combinations of spectral parameters (i.e. 9 HVIs, 12 HVIs, 9 HVIs and 10 PCs, and 12 HVIs and 10 PCs) were applied to maximum likelihood classifications to identify the optimal combinations for vegetation classification. The overall accuracy was 67.5% using 9 HVIs, 71.6% using 12 HVIs, 92.5% using 9 HVIs and 10 PCs, 95.7% using12 HVIs and 10 PCs. The spectral parameter combination of 10 PCs and 12 HVIs was the most powerful for vegetation classification, with an overall accuracy of up to 95.7%. Classification accuracy using other combinations of spectral parameters requires further study.