Image Recognition Of Rock Mass Cracks Based On Deep Learning And Coordinate Extraction

This article mainly takes the granite cracks in the rock mass at the nuclear waste landfill laboratory site in northwest China as the research target.Through the on-site photos of the rock outcrops of the mountains,the improved neural network algorithm is used to segment and analyze the image recognition and pixelization examples.The location of outcrop cracks is roughly identified.After obtaining the approximate location of the cracks,the computer machine vision processing technology is used to perform fine processing to obtain the precise location of the cracks,which lays the foundation for the subsequent 3D modeling of the cracked granite body.The research work and results are summarized as follows:(1)Based on the existing semantic segmentation network framework Deeplab V3+,add the hybrid domain neural network attention mechanism to its encoder network module,and establish the ADeeplab network framework.Compared with the original network framework,it has obtained better recognition performance and has better recognition.Target concentration can effectively eliminate interfering objects that rarely appear in the rock outcrop image.(2)By using a SLR camera to shoot the on-site rock outcrops,the larger-resolution pictures taken back are equally divided into 16 small-resolution photos,and the outcrop photos are marked with fissures to create a training set and a validation set.And use the ADeeplab neural network model to train the marked training set to get the approximate location of the fracture,and at the same time establish a data augmentation module to expand the training set to obtain a mask image of the approximate location of the rock fracture.(3)In order to more finely confirm the specific location of the cracks and reconstruct the cracks,a machine vision-based crack extraction algorithm is proposed.The algorithm uses the method of direct skeleton extraction from the crack mask recognition map,and then optimizes the extracted skeleton.The skeleton pruning algorithm based on scoring elimination system completes the task of pruning the false branches of the skeleton diagram,and only the large correct branches are retained.The large correct branches can roughly describe the specific situation of the outcropping cracks.(4)The key points of the fracture skeleton map are extracted.The key point extraction requirements are the fracture endpoints,the fracture intersections and the key inflection points of the fracture,and their classification is convenient for subsequent reconstruction and merging of the fractures.Then,by extracting the two edges of the key points of the fissure in the skeleton diagram,the connection of the key points is judged,and the fissure map is reconstructed according to the results,and the coordinate of the key points of the fissures of the two-by-two combination is output,and the direction of the fissures in the original image can be successfully reproduced.(5)According to the characteristics such as the trend and slope of the two connected cracks obtained in the previous step,the angle slope method is proposed to merge the two connected cracks in the previous step,output the coordinates of the long crack,and package the program at the same time to establish an intelligent crack recognition system.The subsequent 3D modeling analysis laid the foundation.