The 2017 International Workshop on Advances in SAR: Constellations, Signal processing, and Applications ( ISPRS)

1 / 2017-03-31 12:58:51

Fusion-based deep convolutional neural networks for SAR automatic target recognition

8699,13306,13307,fusion

Recent breakthroughs in algorithms related to deep convolutional neural networks (DCNN) have stimulated the development of various of signal processing approaches in the specific application of Automatic Target Recognition (ATR) using Synthetic Aperture Radar (SAR) data from the MSTAR standard data set. Inspired by the more efficient distributed training such as inception architecture , highway network and its updated version residual network, a brandnew network structure which integrates all the merits of each version is proposed to reduce the data dimensions and the complexity of computation. The detailed procedure presented in this paper consists of the feature fusion to make the representation of SAR images more distinguishable after the extraction of a set of features from different DCNN architectures, followed by a trainable classifier. In particular, the experimental results on the 10-class benchmark data set demonstrate that the presented architecture can largely improve the recognition performance compared with original network design as well as enhance the efficiency of the model.