基于红外热成像的地铁隧道渗漏水病害智能识别

Intelligent Identification of Water Leakage Disease in Subway Tunnel Based on Infrared Thermal Imaging

为了准确、高效检测出隧道渗漏水病害,本文提出利用红外热成像技术采集图像数据,采用结合MobileNet V2的U-Net模型进行地铁隧道渗漏水病害智能识别的方法。结合MobileNet V2的U-Net模型既继承了MobileNet V2的轻量化优势,又保留了U-Net在小样本数据集上良好的分割功能。将从我国南方一地铁隧道内采集到的878张红外热成像图作为原始数据,对数据进行分割标注预处理后,建立标签为地铁隧道渗漏水的数据集。将数据集按9∶1的比例划分为训练集和测试集,使用本文方法进行渗漏水病害识别。结果表明:在保证识别精度的同时,计算参数量仅为原来的1/18,大幅降低了运算量;对小面积点状、中等面积条状和大面积复杂形状三类隧道渗漏水病害均有较好的识别效果;渗漏水区域和周围暗角均为蓝色时,采用传统数字图像处理方法比较容易受到暗角区域影响,而采用本文方法可以较准确识别渗漏水区域,说明本文方法识别效果优于传统数字图像处理方法,值得推广。

In order to detect tunnel water leakage diseases accurately and efficiently,this paper proposed the use of infrared thermal imaging technology to collect image data,and adopted the U-Net model combined with MobileNet V2 for intelligent recognition of subway tunnel water leakage diseases.The U-Net model combined with MobileNet V2 inherits the lightweight advantage of MobileNet V2 while retaining the good segmentation function of U-Net on small sample datasets.878 infrared thermal imaging images collected from a subway tunnel in southern China were used as raw data.After segmentation and labeling preprocessing,a dataset labeled as subway tunnel water leakage was established.The dataset was divided into a training set and a testing set in a ratio of 9∶1,and the method proposed in this paper was used to identify water leakage diseases.The results show that while ensuring recognition accuracy,the number of calculation parameters is only 1/18 of the original,significantly reducing the computational complexity.It has a good identification effect on three types of tunnel water leakage diseases including small area dots,medium area strip,and large area complex shape.When the water leakage area and surrounding dark corners are both blue,traditional digital image processing methods are more susceptible to the dark corner areas.However,the method proposed in this paper can accurately identify the water leakage area,which indicates that the recognition effect of this method is better than that of traditional digital image processing methods,and is worth promoting.