Research On Detection And Identification Of Tunnel Lining Diseases Based On Deep Learning

In tunnel,in use process as time goes on and around the tunnel geological changes and the effect of tunnel lining surface cracks,leakage and loss of diseases,as well as the inside lining will be the basis,imperfect and permeable,and many other diseases,the serious influence the normal operation of the tunnel and the service life,so the tunnel lining timely detected disease epidemic situation is very important.At present,in the field of detection and identification of tunnel lining diseases,there are few methods to detect and identify tunnel lining diseases based on mine penetrating image,the main reason is that it is difficult to manually extract radar images of tunnel diseases,and the recognition rate is low.The deep learning based on convolutional neural network can automatically extract the characteristics of tunnel lining diseases,avoid the problem of information loss in the process of manual feature extraction,and thus improve the accuracy of detection and recognition.Therefore,the deep learning-based detection and recognition of tunnel lining diseases are studied in this paper.The main contents are as follows:(1)Building tunnel lining disease database.The data collected in this paper came from a tunnel design institute in Xi ’an.Through data collection and data amplification processing,a high-quality tunnel lining disease image database was established,and the disease labels made by experts ensured the authenticity and reliability of the data.(2)The identification of tunnel lining diseases based on deep learning is studied.Paper selects the currently popular VGG16 and Res Net34 convolution neural network to build lining disease recognition system,selects the SGD and Adam optimizer to improve the network convergence speed and accuracy,with the help of the GPU parallel computing ability and excellent Tensorflow framework through the contrast test,finally draw Res Net34 model based on Adam optimization algorithm can quickly and efficiently to classification of tunnel lining disease identification,identification results can be achieved 96.15% accuracy.(3)Studying a detection algorithm based on the tunnel lining YOLOv3 disease,using the improved K-Means++ clustering algorithm is more suitable for data size of the prior frame,this paper chooses a priori box and the real boundary box IOU value as the clustering of reference standards,for the convenience of IOU value calculation,moved all box at the center of the coordinate origin,according to the overlap of the horizontal ordinate can be easier to calculate the IOU values.Then,the obtained priori box of new size was used in YOLOv3 algorithm to detect tunnel lining diseases.More small-size lining diseases were detected,and the overall position of diseases selected by the box was more accurate.