Surface Defect Detection Of Hydraulic Tunnel Based On Depth Convolution Neural Network

The concrete lining of the hydraulic tunnel in the infrastructure of the water control project is easy to crack because of the influence of the flow scour,the internal pressure,the change of the surrounding rock and the temperature contraction.Therefore,it is necessary to regularly inspect,evaluate and maintain the defects on the surface of the hydraulic tunnel to ensure the healthy operation of the infrastructure of the water conservancy project.With the development of information and automation,automatic,cost-effective defect detection of tunnel surface has become a research hotspot and focus.This paper focuses on the automatic detection of surface defects of hydraulic tunnel based on convolution neural network.The specific research contents are as follows:First of all,in view of the difficulties in obtaining the environmental image of the hydraulic tunnel and the low quality problems,the image preprocessing is carried out to optimize the data set for the subsequent defect feature extraction.Data augmentation and data enhancement are used to increase the number of images and highlight defect features.At the same time,the depth residual network is introduced to denoise the image,improve the quality of the image,and improve the accuracy of subsequent defect detection.Secondly,aiming at surface defect recognition,this paper proposes a kind of image defect recognition network of hydraulic tunnel based on full convolution network.In this network,convolution and deconvolution are used to predict pixel classes.In the convolution layer,the relu activation function and batch normalization are used to reduce gradient disappearance and enhance feature propagation.At the same time,clique blocks are added after the maximum pooling layer to increase the mutual utilization of low-level features and high-level features and improve the detection accuracy of small defects.In view of the difficulty of building image data sets in practical engineering applications,this paper proposes a two-stage convolution neural network,which is mainly composed of a segmentation network and a decision network.It is suitable for learning defect feature information from a small number of samples and can effectively improve the practicability of defect detection.Then,image processing method is used to extract defect width information and obtain defect quantification information.Finally,the algorithm is verified by experiments.The experimental results show that the method proposed in this paper can detect the defects quickly and accurately,and provide quantitative information,which provides strong data support for the risk assessment and maintenance of the surface defects in the later stage of the hydraulic tunnel,and has practical engineering significance.