DoNet: a light network for SAR ship detection

With increasing maritime traffic, ship detection is vital for safety. SAR enables high-resolution imaging under all weather conditions, but traditional SAR-based methods struggle with small target detection. We propose an enhanced YOLOv8- based network-DoNet, incorporating Dual-Path Feature Fusion Convolution (DPFFC) and Dynamic Cross-Interaction Module (DCIM). DPFFC optimizes feature extraction by reducing initial channels and restructuring features efficiently, while DCIM captures intricate feature interactions via depthwise and pointwise convolutions, nonlinear activation, and dynamic feature fusion. Our approach significantly improves small target detection and computational efficiency, achieving comparable or superior performance to multiple SOTA models with nearly half the parameters, demonstrating its strong potential for maritime monitoring