Selection of Vegetation Indices in Hierarchical Structural Based Vegetation Species Classification

Abstract

A pilot study has been carried out to test the effectiveness and feasibility of a multi-level classification procedure in handling a large number of vegetation species and aims to improve the classification accuracy. In this study, the research question is whether a selected vegetation index has its significance towards a specific vegetation species group. Ten vegetation indices extracted from in-situ hyperspectral remote sensing data has been applied and one-way ANOVA analysis (with significance level, α = 0.01, 0.05, and 0.1) has been carried out to evaluate significant difference of vegetation indices in pair-wise vegetation species. In the multi-level classification procedure, vegetation species were classified continuously from one level to the next level until a good classification result has been achieved. Results indicated that multi-level classification procedure has effectiveness in handling vegetation species where accuracy has been improved from the first classification level to the second level. The study also has highlighted the significance of selected vegetation indices subsets toward different vegetation species in the multi-level classification procedure where the highest accuracy has been given by the selection in dedicated vegetation species groups.