The feasibility analysis of applying NMF in SAR target recognition

Synthetic Aperture Radar automatic target recognition (SAR ATR) plays an important role both in national defense and civil applications. In this paper, the recognition process via Non-negative Matrix Factorization (NMF) is visualized and analyzed to verify the feasible of applying NMF in SAR ATR. As a matrix decomposition method, NMF can obtain part-based representation, and can be applied in SAR ATR as a pattern recognition feature. However, most of researches don´t consider the intrinsic character and the physical meaning of NMF when using it in SAR images. This paper starts from the basic decomposition formulas, explores the physical meaning of the base matrix and the encoding matrix, and obtains the linear relationship between the sample to be recognized and the train samples. The experiments show that NMF can obtain better performance than other pattern features, and can obtain spare representation with practical physical meaning.