高光谱图像亚像元级地物端元提取方法

Endmember extraction method for subpixel materials in hyperspectral imagery

针对高光谱图像中以亚像元形式存在的地物的端元光谱提取问题,提出凸面几何理论和部分非负矩阵分解相结合的端元提取方法.通过去噪的正交基子空间投影方法和相似度比较获得原始图像中的纯像元端元,利用纯像元端元光谱对图像逐点求取丰度和重构误差,对误差大于设定阈值的像素集合进行部分非负矩阵分解,求得亚像元级地物的端元光谱.实验结果表明,该端元提取方法能够弥补传统方法的不足,从而实现对亚像元级地物端元光谱的有效提取.

An endmember extraction algorithm based on the theory of convex geometry and partial nonnegative matrix factorization was proposed in order to solve the problem of extracting the endmembers of subpixel materials in hyperspectral imagery. The dewhitening endmember extraction algorithm based on orthogonal bases of subspace and the similarity comparison were applied to obtain the endmembers containing pure pixels, and the reconstruction error （RE） for every pixel was calculated only with these endmembers. A partial nonnegative matrix factorization algorithm was implemented with the pixels whose REs are greater than the preset threshold, and the endmembers of the subpixel materials were achieved. Experimental results demonstrate that the algorithm can remedy the weakness of the traditional methods and availably retrieve endmembers of the subpixel materials.