Objective Synthetic Aperture Radar (SAR) has established itself as an indispensable tool for marine monitoring due to its all-weather, day-and-night imaging capabilities. However, automatic ship detection in SAR imagery remains a significant challenge, primarily stemming from intense coherent speckle noise, extreme variations in target scales, and complex background interference in inshore scenarios. Existing lightweight models often struggle to maintain high detection precision and robustness under these varying conditions. To address these issues, this paper proposes an improved lightweight detection model, SSD-YOLO, based on the YOLOv8n architecture, designed to enhance performance in complex marine environments.

Method The SSD-YOLO model establishes a hierarchical "Reinforce-Purify-Fuse" synergistic framework by integrating three theoretically grounded innovative modules. First, to address the issue of weak target signals amidst noise, a Parameter-free Target Response Enhancement Module (C2f_SimAM) is embedded in the backbone network. By leveraging an energy-function-based SimAM attention mechanism rooted in neuroscience, this module adaptively calculates 3D attention weights. It effectively enhances the linear separability of target neurons and suppresses background clutter without adding any model parameters. Second, a Lightweight Anti-Noise Feature Aggregation Module (SAR_SPPF) is deployed at the end of the backbone. Distinguished from traditional spatial pyramid pooling, this module utilizes Ghost Convolutions for computational efficiency and introduces a novel dual-path pooling strategy. Theoretical derivation based on the Gamma distribution properties of SAR speckle noise demonstrates that combining small-kernel Average Pooling (to smooth background variance) with large-kernel Max Pooling (to preserve strong scattering peaks) maximizes the signal-to-noise ratio. Third, a Direction-Sensitive Multi-Scale Fusion Module (C2f_DSConv) is introduced in the neck network. This module replaces standard convolutions with direction-sensitive depthwise separable convolutions. By employing specific kernel orientations (0°, 45°, and 90°), it finely captures the anisotropic texture and geometric morphology of ships, significantly improving detection capabilities for densely arranged and multi-scale targets.

Results Extensive experiments were conducted on the SSDD and HRSID datasets. On the SSDD dataset, the model achieved 97.0% precision, 96.4% recall, 99.0% mAP@0.5, and 74.3% mAP@0.5:0.95. Despite these significant performance gains, the model remains highly lightweight, maintaining approximately 3.01M parameters and 7.9G FLOPs. Generalization experiments on the more challenging high-resolution HRSID dataset demonstrated superior robustness, achieving an mAP@0.5 of 94.9%. Quantitative analysis of feature map SNR further validated the internal logic of the proposed synergistic framework.

Conclusion This research provides an efficient and robust solution for SAR ship detection by combining lightweight anti-noise,target enhancement,and direction-sensitive feature capture strategies.