GA-SAR-Det: A Multi-scale YOLOv11 Architecture for Small and Densely Distributed SAR Ship Target Detection

Synthetic Aperture Radar (SAR) ship detection remains challenging due to speckle noise, small target size, and complex maritime backgrounds, particularly in nearshore and densely distributed environments.Existing SAR ship detectors often experience degraded performance on small targets because fine-grained spatial information is progressively lost during feature extraction and multi-scale fusion.This paper proposes GA-SAR-Det, an improved YOLOv11n-based architecture centered on a novel Fibonacci-inspired channel progression strategy for enhancing multi-scale feature representation.Rather than relying on conventional channel scaling, the proposed strategy progressively allocates feature channels according to a Fibonacci recurrence, aiming to preserve feature continuity.To fully exploit the redesigned channel allocation, the proposed architecture integrates C3Ghost modules, a four-scale bidirectional PAN-FPN for enhanced multi-scale feature fusion, and a dedicated P2 detection head for high-resolution small-object localization.In addition, the CIoU regression loss is replaced with PIoUv2 to improve bounding-box optimization stability for small SAR targets.Experiments on the SSDD and HRSID public benchmarks show that GA-SAR-Det achieves mAP@0.5 scores of 98.5% and 91.5%, respectively, outperforming the retrained YOLOv11n baseline under identical experimental settings.These results demonstrate that the proposed Fibonacci-inspired channel progression, together with the supporting architectural enhancements, effectively improves the detection of small and densely distributed ship targets in complex SAR maritime environments.