基于改进YOLOv5的隧道围岩沉降自动识别与实时测量技术研究  

Study on Automatic Identification and Real-time Measurement Technology for Tunnel Surrounding Rock Settlement Based on Improved YOLOv5

作　　者：郝翊杰 李刚[2] 沈丹[3] 邓有为 刘怡阳 HAO Yijie;LI Gang;SHEN Dan;DENG Youwei;LIU Yiyang(School of Electronics and Control Engineering,Chang'an University,Xi'an 710064;School of Energy and Electrical Engineering,Chang'an University,Xi'an 710064;Security Office,Chang'an University,Xi'an 710064)

机构地区：[1]长安大学电子与控制工程学院,西安710064 [2]长安大学能源与电气工程学院,西安710064 [3]长安大学,西安710064

出　　处：《现代隧道技术》2023年第5期58-66,共9页Modern Tunnelling Technology

基　　金：陕西省重点研发计划项目(2023-YBGY-264,2020ZDLGY09-03);广西重点研发计划(桂科AB20159032)。

摘　　要：针对隧道内部环境复杂,施工期间对围岩沉降监测较为困难的问题,提出了一种基于改进YOLOv5的隧道围岩沉降自动识别与测量算法。在隧道施工关键位置安装观测靶标,利用工业相机实时拍摄靶标,进而采用目标检测模型自动识别靶标,通过计算靶标沉降间接测量围岩沉降值,并基于YOLOv5s网络模型,引入ECA-Net通道注意力机制,使网络在通道方面关注待检测靶标,避免通道降维操作,以增强对目标的感知能力。采用改进的α-IoU目标回归损失函数,使得模型总体上对图像特征的学习能力和目标检测精度都有所提升。试验结果表明,改进目标检测模型在工业相机拍摄靶标数据集下的识别精度与召回率分别达到了98.5%和98.4%,检测精度较高。分析误差可知,在拍摄距离0~10 m的情况下,绝对误差在6 mm以内,当距离待测靶标15~25 m远时,绝对误差在15 mm以内,误差较小。In response to the complex internal environment of the tunnel and the difficulty in monitoring surrounding rock settlement during construction,an improved YOLOv5 algorithm for automatic identification and measurement of tunnel surrounding rock settlement is proposed.The observation targets are installed at key locations of tunnel construction,and industrial cameras are used to shoot the targets in real-time,and then the target detection model is used to automatically identify the targets.The settlement of surrounding rock is indirectly measured by calculating the settlement of the targets.Based on the YOLOv5s network model,an ECA-Net channel attention mechanism is introduced to enable the network to focus on the target to be detected in channel aspect,in order to avoid the chan⁃nel dimension reduction and enhance its perception ability for the target.The use of an improvedα-IoU target re⁃gression loss function has improved the learning ability of the model for image features and target detection accuracy on the whole.The experimental results show that the improved target detection model achieves recognition accuracy and recall rate of 98.5%and 98.4%respectively under the industrial camera shooting target dataset,with high detec⁃tion accuracy.By analyzing the error,it can be seen that the absolute error is within 6 mm when shooting at a dis⁃tance of 0~10 meters.When shooting at a distance of 15~25 meters from the target to be tested,the absolute error is within 15 mm,which is relatively small.

关 键 词：隧道围岩 监控量测 目标检测 深度学习 注意力机制 

分 类 号：U452.13[建筑科学—桥梁与隧道工程]