R-Sparse R-CNN: SAR Ship Detection Based on Background-Aware Sparse Learnable Proposals

2025-04-26

AI Summary

To address the challenge of oriented ship detection in complex backgrounds within Synthetic Aperture Radar (SAR) imagery, this paper proposes an end-to-end differentiable detection framework. First, it introduces sparse, learnable Background-Aware Proposals (BAPs), eliminating conventional proposal generation and post-processing pipelines. Second, it designs a Dual Context Pooling (DCP) module to jointly align and model ship and background features within a unified feature space. Third, it incorporates a Transformer-based interaction module to explicitly capture the ternary relationship among ships, their orientations, and contextual background. Evaluated on the SSDD and RSDD-SAR near-shore SAR datasets, the method achieves absolute mAP improvements of 12.8% and 11.9%, respectively, significantly surpassing current state-of-the-art methods. These results validate the effectiveness and generalizability of background-aware sparse modeling for oriented ship detection in SAR imagery.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs

Abstract

We introduce R-Sparse R-CNN, a novel pipeline for oriented ship detection in Synthetic Aperture Radar (SAR) images that leverages sparse learnable proposals enriched with background contextual information, termed background-aware proposals (BAPs). The adoption of sparse proposals streamlines the pipeline by eliminating the need for proposal generators and post-processing for overlapping predictions. The proposed BAPs enrich object representation by integrating ship and background features, allowing the model to learn their contextual relationships for more accurate distinction of ships in complex environments. To complement BAPs, we propose Dual-Context Pooling (DCP), a novel strategy that jointly extracts ship and background features in a single unified operation. This unified design improves efficiency by eliminating redundant computation inherent in separate pooling. Moreover, by ensuring that ship and background features are pooled from the same feature map level, DCP provides aligned features that improve contextual relationship learning. Finally, as a core component of contextual relationship learning in R-Sparse R-CNN, we design a dedicated transformer-based Interaction Module. This module interacts pooled ship and background features with corresponding proposal features and models their relationships. Experimental results show that R-Sparse R-CNN delivers outstanding accuracy, surpassing state-of-the-art models by margins of up to 12.8% and 11.9% on SSDD and RSDD-SAR inshore datasets, respectively. These results demonstrate the effectiveness and competitiveness of R-Sparse R-CNN as a robust framework for oriented ship detection in SAR imagery. The code is available at: www.github.com/ka-mirul/R-Sparse-R-CNN.

Problem

• Detecting oriented ships in SAR images accurately
• Reducing redundant computation in feature extraction
• Improving contextual learning for ship-background distinction

Innovation

• Uses sparse learnable background-aware proposals (BAPs)
• Introduces Dual-Context Pooling (DCP) for unified feature extraction
• Employs transformer-based Interaction Module for contextual learning