To address the challenge of identifying small floating objects in river debris detection, this study proposes an enhanced model that integrates YOLOv8s with Large Separable Kernel Attention (LSKA) and Asymmetric Kernel Convolution (AKConv). The optimized architecture reduces the parameter count of YOLOv8s from 11.1M to 8.3M while improving mAP50 from 68.0% to 77.9% on the "Water Surface Floating Object Dataset" developed by the China Institute of Water Resources and Hydropower Research. Experimental results demonstrate that the proposed method significantly enhances detection accuracy, strengthens occlusion recognition capability, and improves network adaptability to complex river environments. This advancement effectively upgrades the real-time monitoring performance of UAVs in riverine applications, providing an efficient solution for water resource management.
Liu et al. (Wed,) studied this question.