Resolution Progressive Convolutional Neural Network for Fine-Grained SAR Ship Recognition

The advancement in Synthetic Aperture Radar (SAR) imaging quality, coupled with the integration of deep learning techniques, has propelled the research of fine-grained recognition tasks in SAR images. However, recognizing fine-grained ship categories in SAR images is challenging due to discriminative region localization and fine-grained feature learning. In this paper, we propose a resolution progressive convolutional neural network (RP-CNN) that automatically identifies discriminative regions and learns fine-grained features. The model gradually finds salient regions from input and enhances their resolution by localization proposal network (LPN) and attention proposal network (APN), resulting in more accurate recognition and classification. This approach achieves satisfactory results on the FUSAR-Ship dataset.