面向感兴趣类别的约束非负矩阵分解算法

Oriented non-negative matrix factorization algorithm constrained class-of-interest

在处理高光谱数据解混问题中,非负矩阵分解是一种非常有效的方法之一。现有的非负矩阵分解方法一般是针对图中所有地物信息的盲分解,然而实际应用中常常并不需要求取全部地物类别的丰度信息。如果只考虑感兴趣类别,那么其它类别会对其产生不可预测的干扰。针对干扰问题,提出了一种基于最小二乘算法预估计并结合最小距离的约束非负矩阵分解算法(LSMDCNMF)。实验表明,所提出的算法在不忽略非感兴趣类别的情况下,有效地提高了感兴趣类别的解混效果。

In daling with hyperspectral data unmixing, non-negative matrix factorization is an effective method. The existing non-negative matrix factorization algorithms are based on the blind decomposition for all abundant information in the chart. Sometimes it is not needed to calculate all of the abundant informa- tion which belongs to all classified information in the image. It estimates class-of-interest abundant infor- mation, then other materials will be regarded as interference existing in the data. As for this problem, the method based on least squares algorithm estimated for abundant and minimum distance constrained non- negative matrix factorization （LSMDCNMF） is proposed. Experimental results show that the proposed method improves the mixing effect of solution of interest category effectively under the neglect of noniterest category.