GhostNet-YOLOv5s: A Lightweight Fusion Network for Water Surface Target Detection

Water surface target detection stands as one of the most critical tasks in autonomous navigation and aquatic vision applications. To address the challenges of low detection accuracy and slow processing speed caused by complex water environments and small target sizes, this paper proposes a novel water surface target detection algorithm integrating GhostNet with YOLOv5s. A new training dataset named WOSSD is introduced to optimize the improved algorithm for better deployment on unmanned surface vehicle (USV) platforms. Specifically, the GhostNet module is incorporated to reduce the number of parameters and accelerate detection speed. Furthermore, a coordinate attention mechanism is embedded into the backbone network to enhance feature extraction capabilities. A bidirectional feature pyramid network (BiFPN) is implemented in the neck layer to strengthen multi-scale feature fusion between shallow and deep layers, thereby improving small target detection accuracy. Experimental results demonstrate that compared with the baseline YOLOv5s, the GhostNet-YOLOv5s achieves a precision increase from 82.3 % to 87.5 % with a 9.08 % improvement in frames per second (FPS).