Extraction of individual tree crown using hyperspectral image and LiDAR data

Abstract

Tree information such as tree height, tree type, diameter at breast height, and number of trees are critical for effective forest analysis and management. In this regard, this paper presents an individual tree extraction method that uses airborne LiDAR (Light Detection and Ranging) and hyperspectral data. The Support Vector Machine (SVM) classifier was first used to extract tree areas from hyperspectral imagery. Then, Principal Components Analysis (PCA) was applied to the hyperspectral imagery to derive a PCA image consisting of five bands. A segmentation process was performed on the four datasets, which are 1) hyperspectral image, 2) LiDAR-based DCM (Digital Canopy Model), 3) Fusion of PCA image and LiDAR-DCM, and 4) Fusion of PCA image, LiDAR-based DCM and NDVI (Normalized Difference Vegetation Index). Finally, individual tree parameters were estimated based on the segmentation results. The field data were then compared with the final results. The result shows that using fusion data set our method can detect 70% and 92% of reference trees in the forest areas with high and low density, respectively. Conclusively, the tree extraction approach based on the fusion of different data sources provided better results than ones using the single data sources.