Nonparametric Detection of Nonlinearly Mixed Pixels and Endmember Estimation in Hyperspectral Images

摘要: Mixing phenomena in hyperspectral images depend on a variety of factors, such as the resolution observation devices, properties materials, and how these materials interact with incident light scene. Different parametric nonparametric models have been considered to address unmixing problems. The simplest one is linear mixing model. Nevertheless, it has recognized that can also be nonlinear. corresponding nonlinear analysis techniques are necessarily more challenging complex than those employed for unmixing. Within this context, makes sense detect nonlinearly mixed pixels an image prior its analysis, then employ possible technique analyze each pixel. In paper, we propose detecting pixels. detection approach based comparison reconstruction errors using both Gaussian process regression model two combined into statistics which probability density function reasonably approximated. We iterative endmember extraction algorithm combination algorithm. proposed detect-then-unmix strategy, consists extracting endmembers, unmixing, tested synthetic real images.