Abstract: [Purposes] Intelligent monitoring of riverine waste and water pollution is critical for aquatic ecological protection,flood control safety,and smart water conservancy development.However,existing object detection methods suffer from low accuracy and poor robustness in complex riverine scenes due to challenges such as densely distributed small objects,strong water surface reflections,and difficulties in recognizing low-texture targets. [Methods] To address these issues,this paper proposes an enhanced model,FPN-YOLOv9,built upon YOLOv9.Specifically,a lightweight channel-spatial dual attention mechanism is innovatively integrated into the PAN-FPN multi-scale feature fusion module in the Neck layer,enabling dynamic enhancement of salient target features and suppression of background noise,thereby significantly improving detection capability for typical pollutants such as floating debris and abandoned vessels. [Findings] Experimental results show that FPN-YOLOv9 achieves a 53.3% increase in F1-score and a 23.4% improvement in accuracy,with high inference efficiency suitable for real-world deployment. [Conclusions] This study provides an efficient and practical technical solution for intelligent river patrol and dynamic water environment monitoring,while offering novel insights into refined design and optimization of feature fusion mechanisms in remote sensing object detection.

Key words: river waste detection, water pollution monitoring, FPN-YOLOv9, attention-enhanced feature fusion, UAV remote sensing