A Generalization Of P-Linear Mixing Model By Combination Of Two Kinds Of Approximator In Hyperspectral Unmixing

Airborne and spaceborne hyperspectral sensors are widely used for detecting and monitoring information such as human activities and environmental changes. Due to the limitation of spatial resolution, the spectral response is recorded as reflectance of a mixture of materials. In order to extract the abundances of these materials, linear and nonlinear mixture models have been developed. Typically, nonlinear mixing methods have received much attention, for they describe the multiple interactions between materials actually occurring in geometrically and spectrally complex scenes such as urban area and scenes containing water. Accurate description of the complicated interactions in the unmixing model remains a challenge. In this paper, a new approach for nonlinear unmixing is proposed as a combination of two kinds of approximator. It takes advantage of two approximators and makes a better description of the unknown interactions. The proposed method has a trait that can be implemented in a matrix form. Compared with typical models with the same characteristic, the proposed model delivers a better understanding of the imagery. Real data of oil spill on Gulf of Mexico are used to test the model. Experiment results show that unmixing with the proposed model leads to a reconstruction error decrease and an excellent performance in abundance estimation.