结合FDA与NMF的高光谱数据解混方法

Combining Fisher discriminant analysis with nonnegative matrix factorization by perspectral data unmixing

高光谱图像解混是遥感图像处理的重要技术之一.利用非负矩阵分解(NMF)进行高光谱图像解混是近年来发展起来的一种方法.这种解混方法假设光谱具有稳定的光谱特性;但实际上光谱经常是多变的,这个现象影响着解混的精度.为了减小这一影响,首先利用Fisher判别分析(FDA)对高光谱数据进行线性变换,而后利用变换后的高光谱数据提出了一种FDA与NMF相结合的高光谱数据解混方法.实验表明新方法能够有效地提高解混精度与效率.

Hyperspectral unmixing is one of the most important techniques in remotely sensed image processing. Nonnegative matrix factorization（NMF） based hyper spectral unmixing has been developed in recent years. NMF based spectral mixture analysis assumes that the spectrum must have a constant spectral signature. However, spectral variability always exists in practical situations, which reduces the accuracy of mixed pixel decomposition. In order to solve the problem, the spectral data were translated by Fisher discriminant analysis （FDA）, and then using the transformed hyperspectral data a new hyper spectral unmixing method is proposed by combining FDA with NMF. Experiments show that the proposed method can improve both accuracy and efficiency of mixed pixels decomposition.