 This research proposes a transfer learning method for synthetic aperture radar, SAR, target classification using deep convolutional neural networks, CNNs. The limited labeled SAR target data is addressed by making knowledge learned from sufficient unlabeled SAR scene images transferable to labeled SAR target data. An assembled CNN architecture consisting of a classification pathway, a reconstruction pathway, and a feedback bypass is designed. The reconstruction pathway is trained with stacked convolutional autoencoders, SCAE, using a large number of unlabeled SAR scene images. Then, the pre-trained convolutional layers are reused to transfer knowledge to SAR target classification tasks, with feedback bypass introducing the reconstruction loss simultaneously. Experimental results demonstrate that transfer learning leads to better performance in scarce labeled training data, and the additional feedback bypass with reconstruction loss helps boost the capability of classification pathway. This article was authored by Zhong Linghuang, Zhong Xupan, and Bin Lei. We are article.tv, links in the description below.