Complex-Valued Multi-Scale Fully Convolutional Network with Stacked-Dilated Convolution for PolSAR Image Classification

Polarimetric synthetic aperture radar (PolSAR) image classification is a pixel-wise issue, which has become increasingly prevalent in recent years. As variant of the Convolutional Neural Network (CNN), Fully (FCN), designed for pixel-to-pixel tasks, obtained enormous success semantic segmentation. Therefore, effectively using FCN model combined with polarimetric characteristics PolSAR quite promising. This paper proposes novel by adopting complex-valued domain stacked-dilated convolution (CV-SDFCN). Firstly, layer different dilation rates constructed to capture multi-scale features image; meanwhile, sharing weight employed reduce calculation burden. Unfortunately, labeled training samples are usually limited. Then, encoder–decoder structure original reconstructed U-net model. Finally, view significance phase information images, proposed trained rather than real-valued domain. The experiment results show that performance method better several state-of-the-art methods.