水力发电2026，Vol.52Issue(7)：86-93,8.
基于U-Net与河马优化的机器视觉驱动的大坝粗差识别方法
A U-Net and Hippopotamus Optimization-based Machine Vision Method for Gross Error Detection in Dams
摘要
Abstract
To address the limitations of traditional monitoring methods,such as susceptibility to noise,low error identification rates,and insufficient robustness that fail to meet high-precision safety monitoring requirements,a machine vision-based error identification method is proposed.By analyzing the architecture and technical logic of U-Net systems,this approach simulates human visual data perception mechanisms to transform one-dimensional time-series monitoring data into more discriminative high-level semantic features.To reduce the uncertainties caused by hyperparameter combinations and manual post-processing strategies in U-Net models,a Hippo optimization algorithm is incorporated for global optimization,thereby enhancing the model's robustness against hidden coarse errors.Practical engineering validation demonstrates significant improvements in error identification rates compared to conventional machine learning algorithms,particularly showing outstanding performance under multi-source noise interference and nonlinear data scenarios.This innovation provides a novel solution for error identification in dam safety monitoring systems.关键词
大坝安全监测/人工视觉模拟技术/误差识别/河马优化算法/U-Net网络Key words
dam safety monitoring/artificial vision simulation technology/error identification/Hippopotamus optimization algorithm/U-Net network分类
建筑与水利引用本文复制引用
赵鹏,邵晨飞,顾昊,许焱鑫,耿峻,刘顶明,刘勇军,张海龙,汪昌港,童广勤,王一鸣,卢太奇..基于U-Net与河马优化的机器视觉驱动的大坝粗差识别方法[J].水力发电,2026,52(7):86-93,8.基金项目
国家重点研发计划(2024YFC3210700) （2024YFC3210700）
