Polsar Image Classification via Transfer Learning and Fully Convolutional Network

Deep learning and convolution neural networks(CNNs) have recently made advances in polarimetric synthetic aperture radar(PolSAR) image classification. Deep learning methods rely on large amounts of data, but acquiring and labeling PolSAR data is challenging. This paper proposes a complex domain fully convolution network(FCN) based on a new transfer learning method(TF-CVFCN) for PolSAR image classification. First, a large amount of unlabeled data is used to train on Unet to obtain the parameters of the pre-trained model. Then the parameters are migrated to the new FCN network to classify a PolSAR image. In this paper, a new transfer learning method is used, which can be performed on the new model after migrating the parameters of the pre-trained network. Experimental results verify the effectiveness of our proposed method.