Mapping urban land cover from high spatial resolution hyperspectral data: An approach based on simultaneously unmixing similar pixels with jointly sparse spectral mixture analysis

In remote sensing data exploitation, spectral mixture analysis is commonly used to detect land cover materials and their corresponding proportions present in the observed scene. In recent years, high spatial resolution airborne hyperspectral images have shown their potential for deriving accurate land cover maps. In this paper, a new spectral mixture analysis model for mapping urban environments using high spatial resolution airborne hyperspectral data is proposed. First, non-local self-similarity is exploited to partition the scene into groups of similar pixels. The spectral signals of the pixels in each of these groups are assumed to be comprised of the same endmembers, but with different abundance values. Then, the similar pixels in each group are simultaneously unmixed using a jointly sparse spectral mixture analysis method. The proposed method was applied to map land cover in Pavia city, northern Italy, using airborne ROSIS data. An overall classification accuracy of 97.24% was achieved for the Vegetation - Impervious surface - Soil model. Our experimental results demonstrate that the proposed jointly sparse spectral mixture analysis model is well suited for mapping land cover in urban environments using high resolution hyperspectral data.