YOLOv11-LREP: A Lightweight Detection Method for Water-Surface Floating Objects on Inland Waterways Under Low-Light and Reflection Interference

Reliable visual detection of small floating objects on the water surface is a prerequisite for environmental monitoring and clean-up tasks performed by unmanned surface vehicles (USVs) on inland waterways. Such scenes are routinely degraded by low illumination at dawn and dusk, strong specular reflections, ripple-induced clutter, and large object-scale variations, which together cause missed detections, false alarms, and unstable localization. Aiming at these practical challenges, this study conducts a scenario-oriented optimization and experimental validation based on the lightweight YOLOv11n detector. We integrate multiple mature attention mechanisms, regression loss functions and data augmentation strategies to develop an improved scheme, YOLOv11-LREP, for floating object detection. The detailed optimizations are as follows: (i) a Coordinate Attention (CoordAtt) module is inserted at the top of the backbone to enhance positional encoding and highlight obstacle-related semantic regions; (ii) three Efficient Channel Attention (ECA) modules are embedded at the multi-scale fusion nodes of the Neck so that reflection- and ripple-induced spurious channel responses can be suppressed at almost no extra cost; (iii) the Powerful-IoU (PIoU) loss replaces the original regression loss to enforce four-side boundary alignment and stabilize convergence on small, blurred-edge targets; and (iv) a joint low-light and reflection augmentation strategy, together with CutMix region-level mixing, broadens the training distribution along the illumination and occlusion axes. Experiments on the public FloW-Img dataset, split into 1200 training and 800 validation images (2024 instances) and run under a fixed random seed (seed = 0, deterministic = true), show that YOLOv11-LREP attains AP50 = 80.1%, AP50:95 = 38.5%, and AP_S = 24.3% with only 2.84 M parameters and 9.3 GFLOPs. On an NVIDIA RTX 4060 Laptop GPU, the model runs at 3.3 ms total per 640 × 640 image (≈303 FPS), satisfying real-time perception requirements while retaining lightweight deployability. The ablation results indicate that different components contribute differently to localization accuracy, small-object sensitivity, and robustness, and that the final configuration provides a balanced trade-off rather than the best value for every individual metric. A systematic threshold sensitivity analysis (F1 fluctuation < 0.2%) demonstrates the stability of the final model.