HRR Profiles Time-Frequency Non-Negative Sparse Coding for Sar Target Classification

Abstract

Abstract-A new approach to classify synthetic aperture radar (SAR) targets is presented based on high range resolution (HRR) profiles time-frequency matrix non-negative sparse coding (NNSC). Firstly, SAR target images have been converted into HRR profiles. And the non-negative time-frequency matrix for each of the profiles is obtained by using an adaptive Gaussian representation (AGR). Secondly, NNSC is applied to learn target time-frequency basis of the training set. Feature vectors are constructed by projecting each HRR profile time-frequency matrix to low dimensional time-frequency basis space. Finally, the target classification decision is found with support vector machine and nearest neighbor algorithm respectively. To demonstrate the performance of the proposed approach, experiments are performed with Moving and Stationary Target Acquisition and Recognition (MSTAR) public release SAR database. The experimental results support the effectiveness of the proposed technique for SAR target classification.

Unsupervised Deep Discriminative Feature Learning for Aerial Image Classification

“…The non negative of the image input data can better simulate the receptive field behavior of mammals [11]. In 2002, Hoyer proposed NSC algorithm which can not only reflect the statistical characteristics of the data but can also obtain the local feature representation of the interested object [12]. Hoyer uses the method of iterative updating dictionary A and sparse efficient S. First, fixing A, consider the optimization of S and then fixing S, update dictionary A.…”