Tunnel Crack Identification Method And Application Based On Neural Network

As my country’s tunnel construction has entered a rapid period,the demand for daily tunnel inspection has gradually increased,and the method of manual crack detection has been unable to meet the current needs for efficient and accurate detection.Using image processing technology to identify tunnel cracks has become an important development direction of tunnel intelligent detection.On this basis,this thesis designs an automatic detection system for tunnel cracks based on improved Faster RCNN and traditional image processing.The enhancement algorithm is used to improve the contrast of the target area,and the improved Faster RCNN network is built to identify and locate the cracks in the enhanced images.Feature extraction is performed on the identified cracks,thereby realizing automatic crack detection.Firstly,aiming at the low contrast of the image caused by the uneven illumination in the tunnel,an enhancement algorithm is designed to improve the contrast of the target area.Use the Poisson function to obtain the grayscale information between pixels,introduce the grayscale information into the Hessian matrix,build an improved Frangi filter model by solving the Hessian matrix,realize the rough extraction of tunnel crack pixels,and then use the adaptive contrast enhancement algorithm to improve the pixels.The contrast of a point with its neighbors.Experiments show that the regional contrast enhancement algorithm can effectively highlight the crack edge features.Then the enhanced Faster RCNN model is used for crack identification and localization.Because of the characteristics of various crack sizes,the Res2 Net branch structure with richer receptive fields is introduced to replace the Res Net bottleneck structure.Because of the diverse characteristics of cracks,the introduction of variable convolution enables the model to have the ability to change the sampling of feature maps.After the improved Faster RCNN model is trained,the precision,recall,and F1-Score of identifying cracks are 85.35%,96.16%,and 90%,respectively.Experiments show that the improved Faster RCNN model can better meet the needs of tunnel crack detection.Finally,the crack features are extracted according to the model detection results.First of all,the Frangi filter image of the crack is obtained for the crack and the background which are segmented by the Otsu method,the crack breakpoint is repaired and morphological operations suppress the noise,and finally,the crack feature is extracted by the skeleton extraction algorithm and the relevant parameters are calculated.Experiments show that this process can extract the fracture skeleton relatively completely.