摘要
钢筋工程检测存在验收人力资源消耗大和时间成本高的问题,特别是楼板双层钢筋的尺寸验收时,由于上层钢筋覆盖下层钢筋导致测量难度增加,而传统的图像处理方法难以满足测量精度要求,为此提出一种基于Res-AA U-Net的楼板双层钢筋尺寸自动测量方法。该方法对Resnet34进行改进,修剪其网络结构并优化损失函数,用改进的Resnet34代替U-Net的特征提取器,用注意力门机制代替跳跃连接,同时在U-Net底部加入改进ASPP模块,构建包含3355张楼板钢筋图像的数据集,最后利用迁移学习技术加快模型训练速度。结果表明:基于Res-AA U-Net模型的钢筋分割效果优于U-Net、Deeplab v3+、HRNet、PSPNet等经典分割网络,平均交并比、像素精确率和召回率分别达到92.81%、96.02%、94.49%;相较于原U-Net,Res-AA U-Net的钢筋直径测量和钢筋间距测量误差分别减小13.63%、5.82%,测量精度满足钢筋工程验收标准中双层楼板钢筋的验收要求,可有效提升钢筋工程验收效率与智能化水平。
Rebar engineering inspection has the problem of high consumption of human resources and high time cost for acceptance.Especially when the size acceptance of floor slab double-layer rebar,the upper layer of reinforcing bars covers the lower layer of reinforcing bars,leading to increased difficulty in measurement.However,the traditional image processing methods are difficult to meet the requirements of the measurement accuracy.Therefore,an automatic method for measuring the size of double-layer reinforcing bars in floor slabs based on the Res-AA U-Net was proposed.The method improved Resnet34 by pruning its network structure and optimizing the loss function,replaced the feature extractor of U-Net with the improved Resnet34,replaced the jump connection with the attention gate mechanism,and at the same time added the improved ASPP module at the bottom of the U-Net to construct a dataset containing 3355 images of floor rebars,and finally accelerated the training speed of the model by using the transfer learning technique.The results show that the model rebar segmentation based on Res-AA U-Net is better than the classical segmentation networks such as U-Net,Deeplab v3+,HRNet,PSPNet,et al.Key metrics including average intersection over union,pixel accuracy,and recall achieve values of 92.81%,96.02%,and 94.49%,respectively.Compared with the original U-Net,Res-AA U-Net rebar diameter measurement and rebar spacing measurement errors are reduced by 13.63%and 5.82%,respectively,and the measurement accuracy meets the acceptance requirements of double floor rebar in the acceptance standard of rebar engineering,which can effectively improve the efficiency and intelligence of rebar engineering acceptance.
作者
陈婉清
李刚
盛明辉
付相林
陈伟
CHEN Wanqing;LI Gang;SHENG Minghui;FU Xianglin;CHEN Wei(Wuhan Yucheng Jiufang Construction Co.Ltd.,Wuhan 430050,Hubei,China;School of Civil and Environmental Engineering,Changsha University of Science&Technology,Changsha 410114,Hunan,China)
出处
《建筑科学与工程学报》
北大核心
2025年第3期103-114,共12页
Journal of Architecture and Civil Engineering
基金
国家自然科学基金项目(514080634)。
