AMANet: Advancing SAR Ship Detection with Adaptive Multi-Hierarchical Attention Network

Xiaolin Ma,Junkai Cheng,Aihua Li,Yuhua Zhang,Zhilong Lin

2024-01-24

Abstract:Recently, methods based on deep learning have been successfully applied to ship detection for synthetic aperture radar (SAR) images. Despite the development of numerous ship detection methodologies, detecting small and coastal ships remains a significant challenge due to the limited features and clutter in coastal environments. For that, a novel adaptive multi-hierarchical attention module (AMAM) is proposed to learn multi-scale features and adaptively aggregate salient features from various feature layers, even in complex environments. Specifically, we first fuse information from adjacent feature layers to enhance the detection of smaller targets, thereby achieving multi-scale feature enhancement. Then, to filter out the adverse effects of complex backgrounds, we dissect the previously fused multi-level features on the channel, individually excavate the salient regions, and adaptively amalgamate features originating from different channels. Thirdly, we present a novel adaptive multi-hierarchical attention network (AMANet) by embedding the AMAM between the backbone network and the feature pyramid network (FPN). Besides, the AMAM can be readily inserted between different frameworks to improve object detection. Lastly, extensive experiments on two large-scale SAR ship detection datasets demonstrate that our AMANet method is superior to state-of-the-art methods.

Computer Vision and Pattern Recognition,Artificial Intelligence,Machine Learning

What problem does this paper attempt to address?

### Problems the paper attempts to solve

This paper aims to solve the difficult problems of detecting small ships and inshore vessels in Synthetic Aperture Radar (SAR) images. Although deep - learning methods have made remarkable progress in SAR ship detection, in complex environments (such as coastal areas), due to limited features and the influence of background clutter, detecting small and inshore ships remains challenging.

Specifically, the paper proposes a novel Adaptive Multi - level Attention Module (AMAM) to learn multi - scale features and adaptively aggregate salient features from different feature layers, which can work effectively even in complex environments. By fusing the information of adjacent feature layers to enhance the detection of small targets and mining salient regions through channel decomposition, thus reducing the adverse effects of complex backgrounds.

In addition, the paper introduces the Adaptive Multi - level Attention Network (AMANet), embedding AMAM between the backbone network and the Feature Pyramid Network (FPN). Experimental results show that AMANet outperforms the existing state - of - the - art methods on two large - scale SAR ship detection datasets.

### Summary of key issues
1. **Detection of small ships and inshore vessels in complex environments**: Existing methods do not work well when dealing with complex backgrounds in coastal areas, especially when features are limited.
2. **Multi - scale feature fusion**: An effective mechanism is required to fuse multi - scale features to improve the detection ability of ships of different sizes.
3. **Interference of background clutter**: How to reduce the influence of background clutter on detection results, especially in inshore scenarios.

### Solutions
- Propose the AMAM module for learning multi - scale features and adaptively aggregating salient features.
- Enhance the detection of small targets by fusing the information of adjacent feature layers.
- Introduce AMANet and embed AMAM between the backbone network and FPN to improve object detection performance.

These innovations enable AMANet to detect small and inshore ships more accurately in complex SAR images.