Target detection with multiple reflection linear unmixing for hyperspectral remote sensing imagery

Linear mixture model has been widely used for abundance estimation in remotely sensed imagery. Since each pixel in the remote sensing images usually covers several meters on the ground and contains more than one material, its spectrum can be considered as a mixture of all the materials residents in that pixel. Linear mixture model simply assumes this mixture is linear and then estimates the abundance fraction by least square approaches. If the surface is smooth, linear mixture can approximately fit the spectrum. However, if the surface is rough, the multiple reflection effect needs to be considered. In this study, we propose a multiple reflection linear mixture which not only consider single reflection linear mixture, but also includes double and triple reflections. In the model, we also adopt one single factor for the probability of multiple reflections or the roughness. The preliminary result shows the multiple reflection linear mixture can fit the spectrum with less error comparing to traditional linear mixture assumption.