Research On Classification Method Of Surrounding Rock On Tunnel Face Based On Digital Image

At present,China is a big country in tunnel and underground space construction in the world.However,in the process of tunnel construction,manual measurement and recording of tunnel face information are still inseparable.With the rapid development of tunnel mechanization and information construction technology,manual recording of tunnel face information is not conducive to improving tunnel construction efficiency and construction effect.The automatic extraction and recording of tunnel face information by rational use of information technology will help to improve construction efficiency and promote intelligent tunnel construction technology.Therefore,this paper mainly uses machine vision technology and digital image processing technology to carry out the research of tunnel face image and rock image acquisition scheme formulation,tunnel face lithology classification and recognition method,tunnel face structural plane information extraction method and tunnel face surrounding rock classification method and system.The main conclusions are as follows :(1)According to the analysis and verification results of the field image acquisition adaptability test,when the tunnel face image acquisition is carried out in the non-high rock temperature and non-rich water tunnel,the influence of dust,temperature and humidity can be ignored after 20 ～ 30 minutes of normal construction ventilation,but it is necessary to use two supplementary light sources before the image acquisition of the tunnel face and place them in the 1.5 m center line of the symmetrical tunnel and 10 ～ 12 m away from the tunnel face.In the process of rock image acquisition,the amplification function of intelligent equipment should not be used in image shooting to ensure that the rock surface is fresh and as full as possible.Secondly,white light should be used for supplementary light acquisition in the dark tunnel environment.(2)By comparing and analyzing the lithology classification and recognition models established by the existing deep learning network architecture(VGG16,Res Net50 and Mobile Net),it is found that the three models are difficult to balance the relationship between prediction accuracy and model calculation.In order to establish a lithology classification and identification model more suitable for practical engineering,based on the existing deep learning network architecture,a lithology classification and identification model with better comprehensive performance is established by improving the topological structure of the neural network and using the Bayesian method to optimize the hyper-parameters.(3)Considering that the digital image processing methods based on image restoration,image denoising,canny edge detection,mathematical morphology processing,skeleton thinning,eliminating error nodes,and Hough transform line fitting are cumbersome to extract the structural plane information of the palm surface,an automatic extraction model of the structural plane information of the palm surface based on the combination of deep learning pixel difference network edge detection model(Pi Di Net)and digital image processing technology is established to extract the number of structural planes per unit area,the number of structural planes and the average spacing of structural planes.(4)The surrounding rock classification system based on lithology classification and recognition model and structural plane information extraction model of tunnel face is constructed,and the application effect of the system is verified in the Xingshan tunnel of Yixing contact line.The accuracy of surrounding rock classification is 70 %,which can effectively meet the needs of practical engineering.