Defect Detection Of Tunnel GPR Image Based On Neural Network

Tunnel void detection is an important part of tunnel maintenance and traffic operation safety.At present,the interpretation and recognition of images in traditional manual survey rely heavily on the personal experience of engineers,which is subjective and prone to missing detection.With the progress of science and technology,deep learning and artificial intelligence have become brilliant in the field of machine vision.The application of neural network in target detection has become the focus of scientific research.Aiming at the problems of complex shape,large scale,vulnerable to geographical environment and limited computing resources of neural network model in GPR images,this article proposes the guiding anchor mechanism and channel pruning strategy to realize intelligent detection of tunnel diseases based on neural network architecture,and develops a tunnel GPR image void detection system based on python Disease detection software provides technical premise for the deployment of construction site.The main contents of this article are as follows:(1)According to the characteristics of tunnel diseases,this article constructs a CNN network integrated with the guidance anchor mechanism to detect the tunnel void.The network consists of four parts:feature extraction,guide anchor region recommendation network,region of interest pooling and classification regression.The improved guidance anchor suggests that the network can predict the region of interest by learning,and optimize the evaluation criteria to better reflect the coincidence degree between the feature region and the calibration.The experimental results show that the classification accuracy of the improved network is 92.74%,and the trained model has good prediction ability of target data.(2)Combined with the work of two-stage target detection research,through analysis and comparison,this article uses model channel pruning strategy to lighten the model.For each channel,a scaling factor ? is introduced to multiply the output of the channel,and the weights of the training network and these scaling factors are combined.Finally,the channel with small scaling factor is removed directly and the pruned network is fine tuned.Although the detection accuracy of the generated model is slightly reduced,the memory consumption is reduced,and the amount of parameters and calculation is greatly reduced.(3)Based on Python framework technology,a software platform is established which takes guiding anchor and channel pruning as the core technology.The tunnel disease system includes image acquisition,image processing and image application.The image acquisition layer collects the image data input of the geological radar,and after the training and testing of the image processing layer and the lightweight processing,finally,it realizes the display and reporting function of the training results in the application layer.The system not only realizes the intelligent detection of tunnel void disease,but also meets the requirements of deployment in the edge environment with insufficient computing resources,and has practical application value.