FRA-YOLO: Frequency-guided receptive-field adaptation for occluded ship detection in complex nearshore backgrounds

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Detecting ships in complex nearshore scenes is difficult because occlusion, dense interactions, small apparent scale, and structured background textures cause missed detections, false alarms, and inaccurate localization. We propose FRA-YOLO, a compact YOLO11n-based detector that combines frequency-guided detail recalibration with spatial receptive-field allocation. The Dynamic Gated Frequency-High Enhancement (DGFH) module uses low-frequency context to perform bounded recalibration of projected high-frequency residuals before neck fusion. The Dynamic Guided Receptive Field Enhancement (DGRFE) module converts frequency-guided branch probabilities into spatial weights for local, medium, and wide receptive fields. Under public general-to-occlusion transfer, FRA-YOLO achieves 62.28% mAP50, 35.09% mAP75, and 34.53% mAP50:95, exceeding YOLO11n by 4.36, 8.40, and 3.65 percentage points, respectively. On NFVessel, it achieves 59.78% mAP50:95 with 3.21 M parameters at 70.96 FPS and reaches 58.40% mAP50:95 on the complex occlusion subset. Ablation and response analyses show that DGFH provides content-dependent, hierarchy-specific recalibration, while DGRFE spatially redistributes receptive-field contributions. FRA-YOLO therefore improves accuracy in challenging nearshore scenes while retaining a compact, real-time detector.