Fully Convolutional Networks for Automatic Target Recognition from SAR imagery

Abstract

In this work, we proposed a fully convolutional network (FCN) that produces a coarse segmentation for automatic target recognition (ATR) from synthetic aperture radar (SAR) imagery. In order to evaluate the FCN, we trained a deep convolutional neural network (DCNN) that classifies the SAR images from the MSTAR into the 10 classes and we converted the DCNN to the FCN. On the test data, our network has achieved a classification accuracy of 97.9%.

© 2017 IEICE

Reference

Proceedings of the 2017 IEICE Society Conference, B-2-9, Sept. 2017

Link

IEICE info