Object Detection in SAR Via Generative Knowledge Transfer

Abstract

To address the data acquisition and labeling problem for object detection in SAR images, a generative transfer learning framework consists with a knowledge transfer network and a object detection network is proposed. The knowledge transfer network generates pseudo SAR images whose spatial distribution are consistent with labeled optical images and feature distribution are similar to SAR images. These pseudo SAR images are further used to improve generalization performance of convolutional neural network based detection models. Experimental results on SAR SHIP Detection Datasets (SSDD) and AIR-SARShip-1.0 datasets confirm that the pseudo SAR images generated by our method can benefit the final detection prediction even no labeled SAR image is given at the training stage.