LW-CMDANet: A Novel Attention Network for SAR Automatic Target Recognition

Deep learning based synthetic aperture radar automatic target recognition (SAR-ATR) plays an significant role in the military and civilian fields. However, data limitation and large computational cost are still severe challenges in actual application of SAR-ATR. To improve the performance of CNN model with limited data samples in SAR-ATR, this paper proposes a novel multi-domain feature subspaces fusion representation learning method, i.e., a lightweight cascaded multi-domain attention network, namely LW-CMDANet. First, we design a four-layer CNN model to perform hierarchical features representation learning via hinge loss function, which can efficiently alleviate the overfitting problem of CNN model by a non-greedy training style with small dataset. Then, a cascaded multi-domain attention module, based on discrete cosine transform and discrete wavelet transform, is embedded into the previous CNN to further complete the class-specific features extraction from both frequency and wavelet transform domains of the input feature maps. Thus, the multi-domain attention can enhance the features extraction ability of previous non-greedy learning manner, to effectively improve the recognition accuracy of CNN model. Experimental results on small SAR datasets show that our proposed method can achieve better or competitive performance than many current existing state-of-the-art methods in terms of recognition accuracy and computational cost.