A SAR target recognition method with frequency and spatial domain enhancement

Synthetic aperture radar (SAR) image target recognition is a hot issue in remote sensing image application. High accuracy SAR target recognition is truly important in both military and civilian fields. Recently, convolutional neural networks (CNNs) have played an important role in the field of SAR image target recognition, however, most of the existing networks incurs some additional problems such as ignoring the influence of speckle noise on the target recognition process, so the recognition accuracy is low. To cope with these problems, this paper proposes a novel SAR image target recognition method based on CNN with frequency and spatial domain enhancement. First, the image is transformed by Gabor feature descriptor in different frequency directions to generate a plurality of feature maps. Then, the image is spatially enhanced using a Laplace transform. The feature maps which obtained by the frequency and spatial domain enhancement together with the original image are used as the network input. Finally, the improved deep residual network (ResNet) is used to complete the target recognition task. Experimental results demonstrate that the proposed method achieves a state-of-the-art accuracy on the MSTAR dataset.