基于生成对抗网络的盾构隧道渗漏水检测

渗漏水是盾构隧道安全危害最大的病害之一,对盾构隧道渗漏水快速精准的检测,是有效控制及整治盾构隧道渗漏水的基础。现有的渗漏水检测方法在自动化程度方面均取得一定的成效,但存在数据采集效率低、现场采集环境要求高、训练数据样本量大等问题。针对上述问题,文章将移动LiDAR采集的盾构隧道强度图像作为数据源,提出了基于生成对抗网络的盾构隧道渗漏水检测方法,从现有的生成对抗网络V GAN模型出发,在标注少量样本的基础上,建立了Dense块作为编码器,残差块作为解码器的Unet模型作为生成器网络,运用改进的深度残差Unet(Improve ResUnet)作为判别器网络,组成DRUnet IRUnet GAN生成对抗网络用于盾构隧道LiDAR强度图像渗漏水检测。实验结果表明,当输入500张、200张、100张少量样本时,文章构建的DRUnet IRUnet GAN生成对抗网络能够达到优于V GAN的盾构隧道强度图像渗漏水检测效果,表明了所改进的网络具有良好的性能。

Water leakage detection of shield tunnel based on generative adversarial network

Water leakage is one of the most hazardous diseases in shield tunnels,and the rapid and accurate detection of leakage water in shield tunnels is the basis for effective control and rectification of shield tunnel leakage.The existing leakage water detection methods have achieved certain results in terms of automation,but there are problems such as low data collection efficiency,high on site collection environment requirements,and large sample size of training data.To address the above problems,a water leakage detection method for shield tunnels based on generative adversarial network is proposed,which takes the intensity images of shield tunnels collected by mobile LiDAR as the data source.Firstly,starting from the existing V GAN model of the generative adversarial network,based on labelling a small number of samples,the Dense block as encoder is established.Secondly,the Unet model with residual block as decoder is used as a generator network.Finally,the Improved Deep Residual Unet(Improve ResUnet)is applied as a discriminator network to form a DRUnet IRUnet GAN generative adversarial network for water leakage detection of LiDAR intensity images of shield tunnels.The experimental results show that the DRUnet IRUnet GAN generative adversarial network constructed in this paper can achieve the leakage water detection effect of shield tunnel strength image better than V GAN when a small number of samples are of 500,200 and 100 are inputted,indicating that the improved network has good performance.