SAR Automatic Target Recognition Using Transfer Learning Approach

In this paper we propose a new approach for Synthetic Aperture Radar (SAR) automatic target recognition (ATR). One of the main obstacles in SAR ATR is the limited availability of datasets that are used for training. In this paper, a deep learning approach is employed for ATR. The proposed scheme is based on employing a pre-trained convolutional neural network (CNNs) as transfer learning. A pre-trained CNN namely AlexNet is utilized as a feature extractor whereas the output features are used to train a multiclass support vector machine (SVM) classifier. The effectiveness of the proposed framework is verified on a public database where the final result using three target classes attain an accuracy of 99.4%.