| The main objective of this research was to develop new algorithms to automate urban forest inventory at the individual tree level using two emerging remote sensing technologies, LiDAR and hyperspectral sensors. LiDAR data contain 3-Dimensional structure information that can be used to estimate tree height, base height, crown depth, and crown diameter, while hyperspectral data contain rich spectral contents that can be used to discriminate tree species. The synergy of two data sources would allow precision urban forest inventory down to individual trees. Unlike most of the published algorithms that isolate individual trees from a raster surface built from LiDAR data to estimate tree metrics, this study worked directly from the vector LiDAR point cloud data for separating individual trees and estimating tree metrics, in order to generate a better accuracy by preserving the original height values. To effectively discriminate a large number of tree species for urban forests, a neural network based classifier was proposed. This classifier is capable of modeling the characteristics of multiple spectral signatures within each species by an internal unsupervised engine and catching spectral difference between species by an external supervised system. To identify species for each individual tree, the classifier was used to analyze hyperspectral data only at the treetops detected from LiDAR, which can avoid the double-sided illumination, shadow, and mixed pixel problems occurred for the crown level based classification. An additional technique was explored to reconstruct forest scenes by using a 3-D vector-based individual tree visualization model. This technique allows the internal structure of each tree and multi-shape property of many trees in a forest to be characterized. The test results for two study areas from the proposed algorithms and synergy of two data sources were encouraging. Future works should be oriented to the exploration of full potentials of LiDAR data and hyperspectral imagery for urban tree characteristics through data fusion techniques. |
