水力发电2025，Vol.51Issue(12)：45-49,69,6.
基于YOLOv8s的闸门水下淤堵物快速识别模型
A YOLOv8s-Based Rapid Recognition Model for Underwater Siltation at Sluice Projects
摘要
Abstract
Aiming to address the challenge of efficiently identifying underwater siltation,including objects like branches and stones with diverse shapes,a rapid identification model for underwater siltation near gates is proposed.The model is based on the YOLOv8s architecture and integrates FasterNet's efficient feature extraction module along with EMA's multi-scale attention mechanism.It also uses MPDIoU as the location loss function,replacing CIoU,to improve the detection of underwater siltation,such as branches and stones.Engineering case analysis shows that the proposed model achieves a precision of 0.821,a recall of 0.513,and a mAP50 of 0.627,effectively distinguishing underwater debris such as branches and rocks,and an FPS of 120.63 f/s,meeting the real-time requirements for underwater detection.关键词
水闸工程/水下淤堵物/智能探测/YOLOv8Key words
sluice project/underwater siltation/intelligent inspection/YOLOv8分类
建筑与水利引用本文复制引用
贾强强,贾骁男,何旺,朱延涛..基于YOLOv8s的闸门水下淤堵物快速识别模型[J].水力发电,2025,51(12):45-49,69,6.基金项目
国家重点研发计划(2022YFC3005404) （2022YFC3005404）
国家自然科学基金资助项目(52309152,U23B20150) （52309152,U23B20150）
江苏省自然科学基金资助项目(BK20220978) （BK20220978）
国家大坝安全工程技术研究中心开放基金(CX2023B03) （CX2023B03）
