提出一种利用光谱滤波器进行遥感图像混合像元全约束分解的新算法。该算法利用端元光谱中与背景光谱正交的光谱成分构建光谱滤波器, 滤除混合像元中的背景干扰成分, 直接获取信号光谱的丰度。采用该光谱滤波器多次迭代分解, 修正单个混合像元的端元光谱空间, 获取其确切的端元光谱配置, 保证了分解时各端元丰度的非负性, 实现混合像元的全约束分解。多光谱数据仿真实验证明, 与全约束最小二乘法(FCLS)和正交投影(OSP)分解法相比, 该方法虽然在时间方面略逊一点, 但其分解结果与实际结果的相关系数高, 均方根误差小,

The presentation of mixtures not only influences the performance of image classification and target recognition, but also is an obstacle to quantitative analysis of remote sensing images. Therefore, a novel spectrum filter based fully constrained mixture analysis algorithm is proposed in this paper to tackle this problem. The spectrum filter, which could wipe off the background spectrum in a mixed pixel, is firstly proposed to obtain the sum-to-one constrained fractional abundance of mixtures in remote sensing images. Since the precise endmember set of a mixture can be obtained by continually modifying the endmember space when minus abundance exists, the spectrum filter based iterative algorithm is present to realize fully constrained mixture analysis. Experimental analysis based on synthetic multispectral data set demonstrates that the proposed algorithm obviously outperforms the popular Fully Constrained Least Square unmixing (FCLS) algorithm and the Orthogonal Subspace Projection (OSP) algorithm. In addition, the proposed algorithm also achieves very promising performance on real hyperspectral images.