Paper Title
Contrastive Representation Learning for Robust SAR Ship Detection in Maritime Environments

Abstract
Ship detection in Synthetic Aperture Radar (SAR) imagery remains a challenging problem due to severe speckle noise, the small spatial footprint of vessels relative to surrounding sea clutter, and highly dynamic maritime environments. Although recent deep learning–based approaches have achieved notable progress, their performance often degrades under unseen acquisition conditions, sensor variability, or complex ocean backgrounds. To address these challenges, this paper proposes a contrastive representation learning framework for robust SAR ship detection that reduces reliance on extensive manual annotations. By leveraging contrastive learning objectives, the proposed method encourages consistent and discriminative feature representations for ship targets while suppressing background interference across diverse maritime scenes. The learned representations are subsequently utilized for ship detection, enabling stable performance under challenging and variable environmental conditions. In addition, external maritime navigation information is explored as an optional post-detection aid for temporal association and vessel tracking, without influencing the representation learning process. Experimental evaluations on public SAR ship detection benchmarks, including SSDD and HRSID, demonstrate that the proposed approach achieves competitive detection accuracy and improved robustness compared with existing supervised and self-supervised methods. These results demonstrate the effectiveness of contrastive representation learning for practical SAR-based maritime monitoring and surveillance applications. Keywords - Synthetic Aperture Radar (SAR), Ship Detection, Contrastive Learning, Self-Supervised Learning, Robust Feature Representation, Maritime Monitoring, Vessel Tracking.