Research On Highway Tunnel Crack Detection Based On Mask R-CNN

According to statistics from the Ministry of Transport,in recent years,more than 20000 new highway tunnels have been added in China.Due to factors such as temperature,load,earthquake,and time,tunnels have developed varying degrees of cracks.To ensure the safety of vehicles and individuals,preventive testing is necessary.Traditional machine vision based graphic processing methods have problems such as low accuracy,low efficiency,and missed detection in detecting small cracks,With the rapid development of artificial intelligence and big data,deep learning technology is gradually being applied to the research of crack image detection,but its accuracy in identifying small cracks still needs to be improved.This article proposes a method for accurately identifying small cracks in highway tunnels based on Mask R-CNN,which improves and optimizes the model structure to achieve high-precision acquisition of crack area and crack information.Firstly,based on the characteristics of highway tunnel cracks and the algorithm framework of Mask R-CNN,a crack detection scheme including target recognition and size calculation is designed,and the implementation process of Mask R-CNN algorithm is preliminarily planned.At the same time,the traditional image algorithm is combined with deep learning to preset the calculation steps of crack feature size.Secondly,aiming at the problem of complete crack region extraction,the image is preprocessed by block,filtering and augmentation,and then the model is trained by data set,and the results are analyzed and evaluated.By analyzing the principle of defect generation,the corresponding dynamic threshold adjustment mechanism and feature fusion mechanism are introduced to realize the image recognition from correcting the local to improving the whole,and further improve the accuracy of crack recognition.Then,according to the ’ Technical Specification for Highway Tunnel Maintenance ’ and other regulations,the evaluation system of crack risk degree is constructed.Further,the traditional image algorithm is used to extract the crack edge and skeleton,and the orthogonal skeleton line method is used to calculate the crack widthFinally,the unified operation from crack identification to the output of prevention and control measures was completed,and its effectiveness and accuracy were experimentally verified.Aiming at the engineering application of crack detection,and considering the practicality and convenience of actual operation,a software interface for automatic crack detection was designed.