Sar Ship Detection based on Swin Transformer and Feature Enhancement Feature Pyramid Network

TL;DR

This paper introduces a SAR ship detection approach combining Swin Transformer as a backbone with a Feature Enhancement FPN to improve detection accuracy, especially for small ships in complex backgrounds.

Contribution

It proposes integrating Swin Transformer with FEFPN for SAR ship detection, addressing CNN limitations in modeling long-range dependencies and enhancing shallow feature maps.

Findings

Outperforms traditional CNN-based methods on SSDD dataset

Better detection of small ships in complex backgrounds

Enhanced feature maps improve overall detection accuracy

Abstract

With the booming of Convolutional Neural Networks (CNNs), CNNs such as VGG-16 and ResNet-50 widely serve as backbone in SAR ship detection. However, CNN based backbone is hard to model long-range dependencies, and causes the lack of enough high-quality semantic information in feature maps of shallow layers, which leads to poor detection performance in complicated background and small-sized ships cases. To address these problems, we propose a SAR ship detection method based on Swin Transformer and Feature Enhancement Feature Pyramid Network (FEFPN). Swin Transformer serves as backbone to model long-range dependencies and generates hierarchical features maps. FEFPN is proposed to further improve the quality of feature maps by gradually enhancing the semantic information of feature maps at all levels, especially feature maps in shallow layers. Experiments conducted on SAR ship detection dataset (SSDD) reveal the advantage of our proposed methods.