摘要
针对石牌弯道航段表面碍航流态在线监测难题,进行图像目标检测应用研究。现有目标检测算法在流态识别领域应用较少,因此在目标航段采集流态特征数据,并自制表面碍航流态数据集SOFSD。为保障检测速度与精度,基于YOLOv5s框架,结合CA(坐标注意力机制)和BiFPN(双向特征金字塔网络)构建YOLOv5s-CA-BiFPN模型,并基于该模型提出一种新型航道表面碍航流态智能识别方法。试验结果表明:相较于YOLOv5s, YOLOv5s-CA-BiFPN模型在精准率和召回率上分别提升2.3%和0.8%,mAP@0.5提升1.3%和mAP@0.5:0.95降低2.2%,在检测效果和泛化性能上均优于YOLOv5s,有效减少漏检与误检,提升小目标检测能力。基于该方法构建航道表面碍航流态智能识别系统,可为智慧航道建设提供参考。
Aiming at the problem of online monitoring of the obstruction flow state on the Shipai curved section,the application of image target detection is studied.The existing target detection algorithms are rarely used in the field of flow state recognition.Therefore,the flow state feature data are collected in the target section,and the surface obstruction flow state dataset SOFSD is self-made.To ensure the detection speed and accuracy,based on the YOLOv5s framework,combined with the CA(coordinate attention)and the BiFPN(bidrectional feature pyramid network)the YOLOv5s-CA-BiFPN model is constructed,and a new intelligent recognition method for the obstruction flow state on the channel surface is proposed based on the model.The experimental results show that the YOLOv5s-CA-BiFPN model improves the accuracy and recall rate by 2.3%and 0.8%respectively compared with YOLOv5s,and the mAP@0.5 is increased by 1.3%and mAP@0.5:0.95 is decreased by 2.2%.It is superior to YOLOv5s in both detection effect and generalization performance,effectively reducing missed detection and false detection,and improving the small target detection ability.Finally,an intelligent recognition system for obstructive flow on the waterway surface is constructed based on this method,which can provide reference for the construction of smart waterways.
作者
梁锴
王梓鹤
韩越
笪贤楠
李明伟
LIANG Kai;WANG Zihe;HAN Yue;DA Xiannan;LI Mingwei(Three Gorges Navigation Authority,Yichang 443000,China;Harbin Engineering University,Harbin 150000,China)
出处
《水运工程》
2025年第8期195-201,共7页
Port & Waterway Engineering
基金
长江航务管理局重点科技项目(2024-CHKJ-007)
长江三峡通航管理局A类科技项目(KJ2022-02A)。
