OptWake-YOLO: a lightweight and efficient ship wake detection model based on optical remote sensing images

Introduction Ship wakes exhibit more distinctive characteristics than vessels themselves, making wake detection more feasible than direct ship detection. However, challenges persist due to sea surface interference, meteorological conditions, and coastal structures, while practical applications demand lightweight models with fast detection speeds. Methods We propose OptWake-YOLO, a lightweight ship wake detection model with three key innovations: A RepConv-based RCEA module in the Backbone combining efficient layer aggregation with reparameterization to enhance feature extraction. An Adaptive Dynamic Feature Fusion Network (ADFFN) in the Neck integrating channel attention with Dynamic Upsampling (Dysample). A Shared Lightweight Object Detection Head (SLODH) using parameter sharing and Group Normalization. Results Experiments on the SWIM dataset show OptWake-YOLO improves mAP50 by 1.5% (to 93.2%) and mAP50-95 by 2.9% (to 66.5%) compared to YOLOv11n, while reducing parameters by 40.7% (to 1.6M) and computation by 25.8% (to 4.9 GFLOPs), maintaining 303 FPS speed. Discussion The model demonstrates superior performance in complex maritime conditions through: RCEA's multi-branch feature extraction. ADFFN's adaptive multi-scale fusion. SLODH's efficient detection architecture. Ablation studies confirm each component's contribution to balancing accuracy and efficiency for real-time wake detection.