基于深度学习的隧道病害图像检测

Image detection of disease in cross-river tunnel based on deep learning

摘要:

随着我国城市地铁的快速发展，隧道的养护变得越来越重要，传统的人工检测方法不仅效率低、成本高，而且耗时，已经不能满足当今的需求。通过对越江隧道中的电缆通道的病害特征进行研究，提出一种基于深度学习的隧道多病害检测的方法，并提出了一种针对隧道病害检测的残差融合模块网络(Resfmnet)，利用深度学习网络提取图像病害特征并进行病害分类，提高了病害的检测能力，所使用的数据集是通过特种机器人在越江隧道中的电缆通道拍摄的视频获得；实验结果表明所提出的网络显示出更高的准确性和泛化性，对多病害的检测的精度mAP达到0.8914，使得越江隧道检查和监控变得高效、低成本，并最终实现自动化。

Abstract:

With the rapid development of urban subways in our country, the maintenance of tunnels has become more and more important. The traditional manual inspection methods are not only low in efficiency, high in cost, but also time-consuming, which can no longer meet today's needs. By studying the disease characteristics of the cable channel in the cross-river tunnel, a method for detecting multiple diseases in the tunnel based on deep learning is proposed, and a residual fusion module network (Resfmnet) for the detection of tunnel diseases is proposed, using deep learning The network extracts image disease features and performs disease classification, which improves the detection ability of diseases. The data set used is obtained from the video taken by special robots in the cable channel of the cross-river tunnel; The experimental results show that the proposed network shows higher accuracy and generalization, and the accuracy mAP of multi-disease detection reaches 0.8914, which makes the inspection and monitoring of the cross-river tunnel more efficient and low-cost, and finally realizes automation.