A particle swarm optimization algorithm for unmixing the polynomial post-nonlinear mixing model

Spectral unmixing is an important technique for hyperspectral data exploring. Recently the nonlinear unmixing technique which considers the nonlinear mixing terms becomes an important issue of spectral unmixing. Here, we consider a particle swarm optimization technique for nonlinear unmixing. Our motivation is to make a first step to exploit the potential capability of PSO for nonlinear unmixing. The proposed algorithm does not need any prior information concerning about the gradient or hessian matrix. Therefore, it can be easily applied to characterize complex nonlinear mixtures. In addition, it provides a stochastic mechanism that can improve the probability to find a better solution. Furthermore, the experimental results indicate that our algorithm can outperform the traditional algorithm for both synthetic and real hyperspectral data.