提出了满足指数分布的概率模型框架下实现非负矩阵分解的目标函数和相应的算法。同时针对非负矩阵分解方法中的基向量无序性这一特点，将基于互信息的特征选择算法与其结合起来解决了基向量的排序问题。利用目标的先验知识获得有利目标表示的特征向量组，然后用该特征向量组进行滤波，分别获得待测图像的特征图，通过加权的方式将所有的特征图合并为一个总的特征图，最后在特征图上通过选取合适的阈值将目标提取出来。使用MIT林肯实验室ADTS（advanced detection technology sensor）高分辨率机载SAR目标数据进行仿真，结果表明该方法是一种精度较高的目标检测算法。

Non-negative matrix factorization (NMF) method is an effective method which decomposes the nonnegative matrix into two non-negative factor matrices. New iterative formulas of the two non-negative factor matrices are proposed based on the exponent distribution in this paper. The bases acquired by this method are disordered. But the order is very important for target detection. A new NMF approach combined Mutual Information is proposed, which used for the detection of SAR images. In the algorithm, priori knowledge of targets is used to obtain the favorable feature vector groups,the feature maps about the test image are respectively gained by using the feature vector groups. All the feature the maps are weighted into a general characteristic map. Finally, the targets are extracted in the characteristic of the map by choosing a suitable threshold. Experimental results of ADTS target high-resolution airborne SAR data show that this method is effective and feasible.