摘要
早期和及时的缺陷检测对于保证水工混凝土结构的安全运行至关重要。基于深度学习的计算机视觉方法无需复杂的手工特征工程,可以在远程图像中自动判定结构缺陷类别,克服了传统人工视觉劳动强度大、主观性强且易出错的缺点。受此启发,提出了一种基于深度学习的缺陷检测方法,在ResNeXt50网络中引入注意力机制,以自适应地重新校准通道级特征响应,使模型更加关注图像中的缺陷信息,增强特征提取能力。测试结果表明,所提方法可以实现88.0%的F1分数,对于常见混凝土缺陷可以实现较好的分类效果。
Early and timely defect detection is essential to ensure the safe operation of hydraulic concrete structures.The deep learning-based computer vision method does not require complex manual feature engineering,and can automati-cally determine the category of structural defects in remote images,overcoming the shortcomings of traditional manual vi-sion that are labor-intensive,subjective and prone to errors.Inspired by this,this paper proposes a deep learning-based defect detection method,which introduces attention mechanism into the ResNeXt50 network to adaptively recalibrate the channel-level feature responses,so that the model can pay more attention to the defect information in the image and en-hance the feature extraction ability.Test results show that the proposed method can achieve an F1 score of 88.0%,and realize a good classification effect for common concrete defects.
作者
曹国金
苏超
王文君
CAO Guo-jin;SU Chao;WANG Wen-jun(Guangzhou Liuxihe Irrigation District Management Center,Guangzhou 510920,China;College of Water Conservancy and Hydropower Engineering,Hohai University,Nanjing 210024,China)
出处
《水电能源科学》
北大核心
2023年第6期137-141,共5页
Water Resources and Power
关键词
水工混凝土结构
缺陷检测
深度学习
卷积神经网络
注意力机制
hydraulic concrete structure
defect classification
deep learning
convolutional neural network
attention mechanism
