Research On Hyperspectral Image Unmixing Based On Multilinear Mixing Model

In the process of unmixing hyperspectral image,it is difficult for the linear spectral mixing model to describe the spectral mixing phenomenon of hyperspectral image in complex scenes,and the fully constrained nonlinear unmixing for hyperspectral image has such problems as the complex spectral mixing model,the complex mixing objective functions of unmxing and long unmixing time.In this paper,a multilinear unmixing algorithm for hyperspectral image is proposed to solve the problems of complex unmixing objective function and long unmixing time in fully constrained nonlinear unmixing algorithm.First,according to the multilinear spectral mixing model,a fully constrained multilinear unmixing objective function is established.Thus,the multilinear spectral unmixing problem is transformed into the optimal solution problem.Then,by utilizing the [0,1] search domain of the differential search algorithm and the “sum-to-one” boundary control mechanism to meet the abundance constraint conditions,the fully constrained multilinear unmixing objective function can be simplified.Finally,the simplified objective function is iteratively optimized to realize multilinear unmixing for hyperspectral image.The experimental analysis of the algorithm was carried out by using the simulation data set and real hyperspectral image.According to the experiments,compared with the fully constrained nonlinear unmixing algorithm for hyperspectral image,the average unmixing time of the proposed algorithm in this paper was reduced by 14.01%.Moreover,avoiding the complex selection process of the weight value of the abundance constraint term.With regard to the unmixing effect,in general,the proposed algorithm can derive a more desired result.Therefore,the proposed algorithm can reduce the time of fully constrained nonlinear unmixing while ensuring the accuracy of the unmixing,and achieve a better unmixing effect.