基于FCN与视场柱面投影的隧道渗漏水面积检测  

Identification of Seepage Area of Tunnel Based on FCN and Field Projection Model

作　　者：高新闻[1] 简明 李帅青 Gao Xinwen;Jian Ming;Li Shuaiqing(Institute of Mechanical and Electrical Engineering and Automation,Shanghai University,Shanghai 200444,China)

机构地区：[1]上海大学机电工程与自动化学院

出　　处：《计算机测量与控制》2019年第8期44-48,共5页Computer Measurement &Control

基　　金：上海市科技委员会项目(17DZ1204203);上海市科技委员会项目(18DZ1201204)

摘　　要：针对隧道渗漏水病害面积检测中由于复杂环境干扰和隧道几何柱面形状影响而造成较大误差的问题,设计了基于FCN与视场柱面投影算法渗漏水面积检测算法;研制了无人病害巡检车,实现了隧道病害数据的无人采集,通过将FCN处理后的渗漏水病害图片进行视场转换和柱面投影模型的优化,提高了所计算病害面积的准确性;实验结果表明,该算法相比OSTU法、分水岭法和自适应阈值法算法使误检率下降至0.0189,有效提升了隧道渗漏水面积检测的精度。In order to solve the problem of larger error caused by complex environmental interference and tunnel geometry shape in the tunnel leakage area detection,a leaky water area detection algorithm based on FCN and field cylinder projection algorithm was designed.The unmanned disease inspection vehicle was developed to realize the unmanned collection of tunnel disease data.The accuracy of the calculated disease area is improved by performing field-field conversion and cylindrical projection model optimization on the FCN-treated water leakage disease picture.The experimental results show that compared with the OSTU method,the watershed method and the adaptive threshold method,the algorithm reduces the false detection rate to 0.0189,which effectively improves the accuracy of tunnel leakage area detection.

关 键 词：深度学习 视场转换 隧道渗漏水病害检测 柱面投影 全卷积神经网络 语义分割 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]