PolSAR Image Classification Based on Relation Network with SWANet

Deep learning and convolutional neural networks (CNN) have been widely applied in polarimetric synthetic aperture radar (PolSAR) image classification, satisfactory results obtained. However, there is one crucial issue that still has not solved. These methods require abundant labeled samples obtaining the of PolSAR images usually time-consuming labor-intensive. To obtain better classification with fewer samples, a new attention-based 3D residual relation network (3D-ARRN) proposed for image. Firstly, multilayer CNN structure used to extract depth features. Secondly, more important feature information improve results, spatial weighted attention (SWANet) introduced concentrate information, which favorable task. Then, features training test are integrated utilized compute score similarity between samples. Finally, determine category Studies on four different datasets illustrate 3D-ARRN model can achieve higher than other comparison few data.