Polsar Image Classification with Complex-Valued Residual Attention Enhanced U-NET

In this paper, an end-to-end classification method for polarimetric synthetic aperture radar (PoISAR) images through complex-valued residual attention enhanced U-Net is proposed, which incorporates complex-valued operation to utilize phase information and residual attention modules to enhance discriminate features in multiple resolutions. Besides, the deep supervision strategy can not only enhance the ability to learn an effective representation for each scale, but also speed up and stabilize the training process. The experiments clearly demonstrate that our proposed method can achieve state-of-the-art performance compared with recently proposed approaches based on deep belief networks, autoencoders and convolutional neural networks.