SAR图像目标检测的互信息非负矩阵分解算法  

Non-negative matrix factorization based on mutual information for detecting target in SAR images

作　　者：于红芸[1,2] 姜涛[3] 关键[2] 

机构地区：[1]鲁东大学数学与信息学院,烟台264025 [2]海军航空工程学院信息融合研究所,烟台264001 [3]海军航空工程学院兵器科学与技术系,烟台264001

出　　处：《中国图象图形学报》2011年第1期129-134,共6页Journal of Image and Graphics

基　　金：国家自然科学基金项目(60672140);全国优秀博士学位论文作者专项资金(200237);教育部新世纪优秀人才支持计划(NCET-05-0912);鲁东大学校基金(LY20065901)

摘　　要：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M IT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.

关 键 词：非负矩阵分解 互信息 特征选择 目标检测 合成孔径雷达图像 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]