Urban Land Cover Mapping from Airborne Hyperspectral Imagery Using a Fast Jointly Sparse Spectral Mixture Analysis Method

Abstract

Due to the fragmented compositional structure of urban scenes, many pixels are mixtures of multiple materials even in high spatial resolution airborne hyperspectral data. In the past ten years, sparse regression based spectral unmixing methods have achieved some noticeable results. Recently, Chen et al. proposed a jointly sparse spectral mixture analysis model for urban mapping. Their model has a high computational load, however, and wrongly detects a water component in residential areas due to the spectral confusion between water, shadow and other low-albedo land cover materials. In this paper, we propose to exclude water from the spectral mixture analysis in urban scenes. In order to decrease the computational load of Chen et al.’s approach, we propose a fast jointly sparse unmixing method. Our experiments demonstrate that the proposed method obtains a slightly degraded result but has a much lower computational load. It is fourteen times faster than their method, and only requires about one-ninth of the memory. A parallel implementation of the proposed method shows its potential to be applied in practical applications, especially in resource-constrained computational environments.