Decision Tree Classification of Urban Vegetation Using Hyperspectral Imagery

The airborne Pushbroom Hyperspectral Imager (PHI) has been developed by the Shanghai Institute of Technical Physics (SITP). Here we analyse PHI data of Shanghai for the purpose of classifying eight vegetation types found in this urban environment. PHI imagery are first pre-processsed by calibrating to albedo using field spectra and the empirical line method, followed by georeferencing. A classification and regression tree approach is then employed to classify the data, with performance assessed against ‘ground truth’ and the result of a supervised maximum likelihood (SML) classifier run on the same data. Compared to the SML approach, slightly improved accuracies are obtained via the boosted decision tree method with principal component analysis. Results indicate the potential of the decision tree classification algorithm to separate urban vegetation types in hyperspectral datasets.