传统图像处理算法在定位衬砌背后的空洞时,往往受到复杂环境因素的干扰,从而降低分割的准确率。为提升分割准确率,对UNet网络模型增加注意力门结构进行改进研究,并与其他主流分割网络进行对比实验。开展空洞的探地雷达二维正演模拟、模型试验和衬砌现场检测工作,为改进UNet网络提供数据支撑。结果表明:在衬砌背后空洞数据集上经过100轮训练后,改进后的AGS-UNet网络相比改进前在准确率(CPA)、交并比(Iou)和平均交并比(MIou)3个评价指标上均提高了3%左右。证明了改进后的AGS-UNet网络能够提高衬砌背后空洞图像分割的精确度,是一种有效的改进网络模型。
Traditional image segmentation algorithms are highly susceptible to the influence of complex environmental factors when it comes to accurately locating cavities behind lining structures. To enhance the accuracy of this segmentation process, an attention gate structure has been incorporated into the UNet network model. Comparative analysis has been conducted, comparing the performance of this modified model with other mainstream segmentation networks. 2-D GPR forward modeling of cavity, model tests of different surrounding rock grades, and field lining detection are carried out. The work includes establishing 2-D models of circular, rectangular and triangular cavities behind tunnel lining by using finite difference time domain method, and obtaining GPR forward images of lining with different shapes of cavities behind lining; Concrete models simulating the existence of voids behind lining under different surrounding rock grades are made, and GPR is applied to detect the voids in the model. Based on an example of a subway tunnel, GPR is used to detect the tunnel lining, analyze the void situation behind lining, and invert the GPR images of voids behind lining. These efforts provide support for improving UNet in many ways. The results obtained from these experiments reveal that, after undergoing 100 rounds of training on the dataset specifically focused on cavities behind lining structures, the improved AGS-UNet network exhibits the best performance in terms of MIou and CPA index coefficients. Comparative analysis demonstrates that the improved AGS-UNet network surpasses the performance of both the FCN network and the classical UNet network, leading to enhanced accuracy in segmenting cavity images behind the lining structures. This improved network model serves as an effective tool and holds significant guiding implications for the rapid identification and quantitative analysis of cavity defects behind lining structures in engineering projects.By enabling prompt and accurate assessment of these cavity defects, this approach contributes to the overall safety and reliability of engineering endeavors.
[1] 肖都,李文杰,郭鹏.基于GPRMax的隧道衬砌检测数值模拟及应用[J].物探与化探,2015,39(4):855-859.
[2] 伍浩.探地雷达技术进行隧道衬砌质量检测的机理及精度分析[D].长沙:长沙理工大学,2013.
[3] 刘杰,张千里,杜翠,等.含缺陷高铁隧道衬砌探地雷达检测模型试验[J].中国铁道科学,2021,42(5):103-111.
[4] 杨艳青,高永涛,贺少辉,等.隧道衬砌背后积水地质雷达检测模型试验研究[J].中国铁道科学,2014,35(6):63-69.
[5] 周陈婴.隧道衬砌检测探地雷达图像分析与工程应用[J].山东大学学报(工学版),2018,48(4):61-68.
[6] 刘宗辉,吴恒,周东,等.频谱反演法在探地雷达隧道衬砌检测中的应用研究[J].岩土工程学报,2015,37(4):711-717.
[7] 钟世航,王荣.探地雷达检测隧道衬砌中的几个问题[J].物探与化探, 2002(5): 403-406.
[8] 李兴. 隊道衬砌病害的探地雷达识别研究及应用[D].大连: 大连理工大学, 2012.
[9] 杜良,刘树才,梁棋念, 等.隧道检测中基本形状空洞探地雷达图像特征[J].地球物理学进展,2015,30(6):2892-2896.
[10] 应国刚. 衬砌背后空洞对隧道结构体系安全性的影响机理研究[D].北京: 北京交通大学, 2016.
[11] 彭跃, 王桂林, 张永兴, 等.衬砌背后空洞对在役隧道结构安全性影响研究[J].地下空间与工程学报, 2008, 4(6):1101-1104.
[12] 李建旺, 曹明星, 周喻. 高速公路隧道穿越富水空洞区协同施工法研究[J]. 地下空间与工程学报, 2020, 16(增2): 798-803.
[13] 毛宁,吕擎峰,郑立晟, 等.基于神经网络的探地雷达隧道衬砌检测方法[J].地下空间与工程学报,2023,19(增1):323-328.
[14] Liu H, Lin C, Cui J, et al. Detection and localization of rebar in concrete by deep learning using ground penetrating radar[J]. Automation in Construction, 2020, 118: 103279.
[15] Tong Z, Gao J, Zhang H. Recognition, location, measurement, and 3D reconstruction of concealed cracks using convolutional neural networks[J]. Construction and Building Materials, 2017, 146: 775-787.
[16] 柳厚祥, 李汪石, 查焕奕, 等. 基于深度学习技术的公路隧道围岩分级方法[J]. 岩土工程学报, 2018, 40(10): 1809-1817.
[17] Xie L H, Zhao Q, Ma C G, et al. U-Net: Deep-learning schemes for ground penetrating radar data inversion[J]. Journal of Environmental and Engineering Geophysics,2020,25(2): 287-292.
[18] Yee K S. Numerical solution of initial boundary value problems involving maxwell's equations in isotropic media[J]. IEEE Transactions on Antennas & Propagation, 1966, 14(5):302-307.
[19] 黎杨,程莉. FDTD算法与Matlab仿真在电磁场与电磁波教学中的应用[J].高师理科学刊,2022,42(11):86-90.
[20] 彭晓迪. 基于改进UNet网络的高分二号遥感影像分类研究[D].北京: 中国地质大学,2020.
[21] 潘晓妍. 基于UNet深度神经网络的颈动脉分叉分割研究[D].杭州: 浙江大学,2019.
[22] Long J, Shelhamer E, Darrell T. Fully convolutional networks for semantic segmentation[A] // 2015 IEEE Conference on Computer Vision and Pattern Recognition[C]. Boston, 2015: 7-12.
