Research And Realization Of Automatic Detection System For Tunnel Lining Cracks

In recent years,with the vigorous development of my country’s transportation industry,a large number of tunnels have been built across the country.Due to the complex external environment of the tunnel,and after years of operation,a large number of linings have appeared many kinds of diseases.Among them,lining cracks cause the most ser Io Us damage to the tunnel,and the detection of lining cracks is also the main task of the daily maintenance of the tunnel.At present,manual inspection is the main method of tunnel lining crack inspection.However,the working conditions in the tunnel are harsh,and the manual inspection has high intensity and low efficiency.With the development of artificial intelligence in computer vision,the application of deep learning in the detection of tunnel lining cracks has become a research hotspot.This paper first establishes a data set of tunnel lining crack images,and through the comparative analysis of experimental results,perfects the classification of the data set,and finally establishes 6000 data sets,which are divided into 11 categories.Using the classic Faster R-CNN detection algorithm to test this data set,the recognition accuracy rate reaches 87.24%.In this paper,a cascade detection framework is designed based on the local similarity of crack images.The first level uses Faster R-CNN detection algorithm,and the second level uses a fully connected classifier.The ratio and confidence level after the first level detection are used as the input of the second level network.Both the ratio and the confidence level reflect the similar relationship between the part and the whole.Experimental results show that this method effectively improves the detection effect of identifying small cracks in high-resolution images.Finally,this paper designs and implements an automatic detection software for tunnel lining cracks.