基于非负矩阵分解的SAR图像目标识别

 特征提取是合成孔径雷达自动目标识别的关键技术，同时也是难点问题之一。本文提出了一种基于非负矩阵分解算法与Fisher线性判别方法的合成孔径雷达图像目标识别的方法，通过基于基向量非负加权组合的形式构建SAR目标图像，能充分利用目标的局部空间结构信息提取目标特征信息实现目标识别。首先将水平集分割预处理后的SAR目标图像样本构成初始矩阵，然后利用非负矩阵分解后得到的权向量作为目标图像的特征向量，再通过依据Fisher线性判别构成的分类器，实现对MSTAR数据中3类目标的识别，并与目前已有的几种典型方案进行对比。试验结果表明该方法是可行且有效的，并能够明显提高对目标识别的稳定性和正确率。

Non-Negative Matrix Factorization For Target Recognition

The feature extraction is one of the key steps and difficulties for Synthetic Aperture Radar(SAR) Auto Target Recognition. This paper proposes a novel method based on Non-negative Matrix Factorization for SAR images feature extraction and target recognition. In order to make full use of local spatial structure information for target feature extraction to achieve target recognition, it takes the form of non-negative weighted combination of basis vectors to construct SAR target images. First, the level set SAR image segmentation method is adopted to get the target image from noisy SAR image, then, after non-negative matrix factorization, the resulting weighted vectors are regarded as the feature vectors of the target images, and finally, Fisher Linear Discriminant is considered as a classifier to perform target recognition. The method is used for recognizing three-type target motels in MSTAR database. Compared to other classical methods, the experimental results show that the new method is an effective approach for SAR images feature extraction and target recognition.