Chromatic intelligence: fusion of colour space transformations for adaptive land-cover classification using SAR imagery

The colours of image pixels can be described within various colour spaces, each designed to emphasize particular properties of the target. However, the effective application of colour spaces for polarimetric SAR (Synthetic Aperture Radar) data in various applications such as land-cover classification is still less explored, although it has the potential to differentiate several land-cover classes more efficiently. Therefore, in this work a critical assessment of different colour‑space models on PALSAR (Phased Array L-band Synthetic Aperture Radar) data has been performed, and based on that a land-cover classification approach has been proposed as a novel application of colour‑space models. This study concentrates on creating pseudo-colour images from fully polarimetric SAR data and deriving various components from different colour‑space models. The criterion for classification are based on colour component statistical measures, which make the algorithm robust and adaptive. The thresholding expression for each class contains an unknown variable whose relationship with the overall accuracy (OA) and false alarm rate (FAR) of that particular class has been established. The obtained relationship has been further optimized to obtain the optimum values of the unknown variables. The effectiveness of the proposed algorithm has been assessed using fully polarimetric PALSAR data, yielding highly satisfactory outcomes.