基于深度学习的公路隧道表观病害智能识别研究现状与展望

Review and prospect of intelligent identification of apparent diseases in highway tunnels based on deep learning

公路隧道服役过程中会产生诸多衬砌病害,其会影响隧道的结构耐久性与运营安全性,对隧道表观病害进行高效智能化识别至关重要。常用的人工巡检方式效率低下且准确率低,而基于深度学习算法进行表观病害智能识别能提高检测的效率和准确性,相较于传统方法而言在实际隧道工程中具有更好的应用前景。利用深度学习可以学习隧道病害的特征信息,有利于未来隧道病害识别智能化的发展。简述深度学习在隧道表观病害识别中的应用原理,从人工拍照方法、数字图像采集和激光扫描技术三方面介绍病害图像的采集,从标注软件和数据增强方法总结数据集的构建和扩充方法,在图像分类、目标检测、语义分割三方面总结深度学习算法在隧道病害检测的应用现状,归纳当前应用的不足之处,最后分析与展望深度学习在隧道表观病害智能化识别方向广泛应用需要研究的问题与方向。

During the service of highway tunnels,many lining diseases will occur,which will affect the structural durability and operational safety of the tunnel.It is very important to identify the apparent diseases of the tunnel efficiently and intelligently.The commonly used manual inspection methods have low efficiency and low accuracy,and the intelligent identification of apparent diseases based on deep learning algorithms can improve the efficiency and accuracy of detection,and has better prospect in actual tunnel engineering compared with traditional methods.The feature information of tunnel diseases can be learned by using deep learning,which is beneficial to the development of intelligent identification of tunnel diseases in the future.This paper briefly describes the principle of deep learning applied in the identification of tunnel apparent disease,introduces the collection of disease images from the three aspects of manual photography method,digital image acquisition and laser scanning technology,summarizes the construction and expansion methods of datasets from the labeling software and data enhancement methods,summarizes the state-of-the-art of deep learning algorithms in tunnel disease detection in three aspects:image classification,target detection,and semantic segmentation,and summarizes the shortcomings of current applications,and finally analyzes and prospects the problems and directions that need to be studied in the wide application of deep learning in the direction of intelligent identification of tunnel apparent diseases.