Tree Species Recognition in Species Rich Area Using UAV-Borne Hyperspectral Imagery and Stereo-Photogrammetric Point Cloud

Abstract

Recognition of tree species and geospatial information of tree species composition is essential for forest management. In this study we test tree species recognition using hyperspectral imagery from VNIR and SWIR camera sensors in combination with 3D photogrammetric canopy surface model based on RGB camera stereo-imagery. An arboretum forest with a high number of tree species was used as a test area. The imagery was acquired from the test area using UAV-borne cameras. Hyperspectral imagery was calibrated for providing a radiometrically corrected reflectance mosaic, which was tested along with the original uncalibrated imagery. Alternative estimators were tested for predicting tree species and genus, as well as for selecting an optimal set of remote sensing features for this task. All tested estimators gave similar trend in the results: the calibrated reflectance values performed better in predicting tree species and genus compared to uncorrected hyperspectral pixel values. Furthermore, the combination of VNIR, SWIR and 3D features performed better than any of the data sets individually, with calibrated reflectances and original pixel values alike. The highest proportion of correctly classified trees was achieved using calibrated reflectance features from VNIR and SWIR imagery together with 3D point cloud features: 0.823 for tree species and 0.869 for tree genus.