High-order Nonlinear Unmixing Algorithm For Hyperspectral Imagery

With the development of economy and society,urban environment and natural environment become more and more complex in composition and spatial structure.At the same time,the accuracy requirement of remote sensing information mining is also improving.This puts forward new challenges for the study of hyperspectral unmixing.The traditional linear spectral unmixing cannot satisfy the accuracy requirement and nonlinear spectral mixing method has become a hot spot of research.With the improvement of the requirement of unmixing,the nonlinear effect of higher order is getting more and more attention in the field of nonlinear unmixing.In order to describe the higher order mixing effect in the scene,the complexity of the spectral mixture model increases,and the parameter size of the model increases,for example the p-Linear mixing model.However,at the same time of achieving complete expression ability,the high-order spectral mixture model also brings difficulties to the inversion of abundance.In this context,this paper mainly discusses the problems that will be encountered in the inversion of abundance based on the high-order spectral mixture model,and proposes corresponding solutions;Besides,in order to express the nonlinear mixture in the field of view,a new high order nonlinear spectral mixture model is proposed.A brief introduction of these algorithms are shown as follows:1.Normalized p-linear algorithm?NPLA?.2.Integrating Spatial Information in the Normalized P-Linear Algorithm?SSDP-NPLA?.3.Multi-Harmonic Postnonlinear Mixing Model?MHPNMM?.First,two nonlinear hyperspectral unmixing algorithms are proposed to improve the accuracy and robustness of high-order nonlinear spectral mixture models.Based on the p-Linear model,the emphasis of this paper is to add prior information in the unmixing to reduce the over fitting of the model.NPLA algorithm takes7)₂ norm constraints directly on the linear and nonlinear parameters of the model,and brings the problem into the convex optimization framework to solve it.The SSDP-NPLA algorithm takes the strategy of local parameter sharing,taking into account the spatial correlation within the image.Then,a new hybrid model of high order spectrum is proposed in this paper to improve the expression ability of the model to the spectral mixture in the scene.The MHPNMM model uses nonlinear mapping to model the local compact mixture,and assumes that the local tight mixing in the field of view is caused by multiple scattering superposition.The experimental results show that the three methods can improve the accuracy of the hyper spectral mixture pixel decomposition.