Mixed pixel in hyperspectral image is actually a nonlinear spectral mixing,which is caused by multiple re-flecting and scattering.Because traditional spectral demixing algorithms are based on linear spectral mixture model, they have low demixing precision.A nonlinear spectral demixing algorithm based on isometric mapping and spatial in-formation is presented under nonlinear mixing assumptions.The algorithm reduces original data into a low-dimensional space by isometric mapping,and spatial information is used to extract endmembers from low-dimensional space.The a-bundances of the obtained endmembers are calculated by full-constrained least squares.Experimental results on real hyperspectral data demonstrate that the proposed approach outperformed N-FINDR method and the geodesic simplex volume maximization(GSVM).%由于多重反射和散射,高光谱图像中的混合像元实际上是非线性光谱混合。传统的光谱解混算法是以线性光谱混合模型为基础,因此解混精度不高。本文在光谱非线性混合模型的基础上,提出一种将等距映射与空间信息结合的非线性光谱解混算法。该算法通过等距映射算法将原始高光谱数据非线性降维到低维空间,并结合空间信息实现端元提取。得到的端元采用全约束的最小二乘法计算相应丰度。真实高光谱遥感数据实验结果表明,采用该算法得到的结果优于 N-FINDR 算法和基于测地线距离的最大单形体体积(GSVM)算法。
展开▼