基于稀疏约束的SAR目标特征提取方法研究

SAR Target Feature Extraction Based on Sparse Constraint Nonnegative Matrix Factorization

特征提取是合成孔径雷达目标识别关键技术与核心任务。为了更好地提取目标特征,稀疏约束将被添加在非负矩阵分解法中,并应用于图像目标特征提取,通过利用稀疏约束的非负矩阵分解方法对SAR目标图像进行分解,构建具有稀疏性的目标特征矢量,提高了特征矢量的类内相似性与类间差异性。利用基于支持向量机的分类方法对MSTAR数据进行目标识别试验,试验结果表明,添加稀疏约束的NMF方法与PCA、ICA以及一般NMF特征提取方法相比,能够显著提高目标识别的稳定性和准确率。

Feature extraction is a key technology and the core task in synthetic aperture radar（SAR） tar get recognition. A non-negative matrix decomposition of the image target feature extraction method based on sparseness constraint is presented in this paper. In order to improve the similarity within the class and the differences between the classes of feature vectors, the SAR target image is decomposed by use of sparseness constraint non-negative matrix factorization method to build sparse target feature vector. MSTAR data are used for target identification test based on support vector machine classification. The results show that the proposed method improves the stability and accuracy of target recognition significantly compared with PCA, ICA and general NMF characteristics extraction methods.