摘要
恶劣地质环境下的川藏铁路高原隧道存在高地应力、高地温、高水压等复杂干扰因素,为实现高原隧道裂缝病害的高效准确定位与评价,提出了基于深度学习的隧道裂缝智能识别及评价方法。首先,通过Mask算法和CLAHE算法对高原隧道裂缝图像预处理;其次,开展基于改进YOLOv5目标检测算法的裂缝定位识别与基于ATT-Unet语义分割算法的裂缝像素分割;再次,通过提取裂缝形态特征信息,采用优化后的最短距离法定量计算裂缝的宽度与长度;最后,建立和扩容不同破坏程度的多场景隧道裂缝图像数据集,进而构建深度学习模型评价体系。依托实际高原隧道工程进行检测对比试验,该智能识别及评价技术的裂缝查准率可达86%,查全率可达96%,调和平均F1值可达91%,交并比可达84%,验证了在高原复杂因素干扰下技术的有效性与可靠性,满足高原隧道检测实时性和准确性的要求。
There are complex interference factors such as high ground stress,high ground temperature and high water pressure in the plateau tunnel of the Sichuan-Tibet Railway under harsh geological environment.In order to realize efficient and accurate location and evaluation of the fracture disease in the plateau tunnel,an intelligent recognition and evaluation method based on deep learning is proposed.Firstly,the crack image of plateau tunnel is preprocessed by Mask algorithm and CLAHE algorithm.Secondly,crack localization recognition based on improved YOLOv5 target detection algorithm and crack pixel segmentation based on ATT-Unet semantic segmentation algorithm are carried out.Thirdly,the width and length of the crack are quantitatively calculated by the optimized shortest distance method by extracting the morphological characteristics of the crack.Finally,multi-scene tunnel crack image data sets with different damage degrees are established and expanded,and then a deep learning model evaluation system is constructed.Based on the actual detection and comparison test of the plateau tunnel project,the crack accuracy rate of the intelligent identification and evaluation technology can reach 86%,the recall rate can reach 96%,the harmonic average F1 value can reach 91%,and the intersection ratio can reach 84%,which verifies the effectiveness and reliability of the technology under the interference of complex factors on the plateau,and meets the requirements of real-time and accuracy of the plateau tunnel detection.
作者
左迪
王鑫
赵丹
ZUO Di;WANG Xin;ZHAO Dan(Tianshui Normal University,Tianshui,Gansu 741000,China)
出处
《黑龙江交通科技》
2024年第5期115-120,共6页
Communications Science and Technology Heilongjiang
基金
国家自然科学基金项目(52068063)
甘肃省高等学校创新能力提升项目(2019A-099)。
关键词
隧道裂缝
目标检测
语义分割
深度学习
tunnel crack
object detection
semantic segmentation
deep learning
