摘要
针对小型土石坝的异常人员和渗漏安全检测效率低、成本高的问题,提出了一种创新的智能巡检方法。该方法基于无人机图像和深度学习目标检测技术,改进了YOLOv8s网络结构,引入细粒度卷积和新型MPDIoU损失函数,细粒度卷积能够捕捉到更丰富的图像细节,而新型的MPDIoU损失函数则有助于提高模型的定位精度。在自制的大坝异常人员和渗漏数据集上进行了实验,实验结果表明,所提出的智能识别算法在人员与渗漏的综合识别能力上具有显著优势,提高了对土石坝工程库区入侵人员和大坝渗漏的检测准确度,并能适应不同环境下的实际应用需求。
In response to the low efficiency and high cost of abnormal personnel and seepage safety detection for small earth and rock dams,an innovative intelligent inspection method is proposed.This method is based on UAV images and deep learning object detection technology,improves the YOLOv8 network structure,and introduces fine-grained convolution and a new MPDIoU loss function.Fine-grained convolution can capture richer image details,while the new MPDIoU loss function helps to improve the positioning accuracy of the model.Experiments were conducted on a self-made dam abnormal personnel and seepage dataset.The experimental results prove that the proposed intelligent recognition algorithm has significant advantages in the comprehensive recognition of personnel and seepage,improves the detection precision of intruders in the reservoir area of the earth and rock dam project and dam seepage,and can adapt to the actual application needs in different environments.
作者
王勃渊
康飞
梁学文
吴卫熊
朱思思
WANG Bo-yuan;KANG Fei;LIANG Xue-wen;WU Wei-xiong;ZHU Si-si(School of Infrastructure Engineering,Dalian University of Technology,Dalian 116024,Liaoning Province,China;Guangxi Zhuang Autonomous Region Water Conservancy Research Institute,Nanning 530023,Guangxi,China;China Yangtze Power Co Ltd,Yichang 443000,Hubei Province,China)
出处
《中国农村水利水电》
北大核心
2025年第2期213-218,共6页
China Rural Water and Hydropower
基金
国家重点研发计划课题资助项目(2022YFB4703404)
广西重点研发计划项目(桂科AB24010003)
国家自然科学基金项目(52079022)。
作者简介
王勃渊(2000-),男,硕士研究生,主要从事大坝安全监测分析研究。E-mail:wangboyuan0701@163.com。;通讯作者:康飞(1982-),男,教授,博士生导师,主要从事大坝安全监测研究。E-mali:kangfei@dlut.edu.cn。
