Unmixing prior to supervised classification of urban hyperspectral images

Abstract

Supervised classification of urban hyperspectral images is a very challenging task due to the generally unfavorable ratio between the number of spectral bands and the number of training samples available a priori, which results in the Hughes phenomenon. Training samples are particularly challenging to be collected in urban environments. A possible solution is to reduce the dimensionality of the data to the right subspace without losing the original information that allows for the separation of classes. In this paper, we propose a new strategy for feature extraction prior to supervised classification of urban hyperspectral data which is based on spectral unmixing concepts. The proposed strategy includes the sub-pixel information that can be obtained with spectral unmixing techniques into the classification process, and does not penalize classes which are not relevant in terms of variance or signal-to-noise ratio (SNR) as it is the case with other transformations such as principal component analysis (PCA) or the minimum noise fraction (MNF). Experiments using urban hyperspectral image data collected by the reflective optics spectrographic imaging system (ROSIS) over the city of Pavia in Italy are discussed, using the support vector machine (SVM) classifier as a baseline for demonstration purposes.