Research On Fracture Segmentation Method Of Rock Mass Drilling Images Based On Deep Learnin

The development of cracks in boreholes is closely related to the quality of rock mass engineering.In the process of modern engineering exploration,digital borehole images taken by borehole TV are used to observe the development of cracks in boreholes.It is very important to identify the crack information of digital borehole image of rock mass quickly and accurately.At present,to solve this problem,most studies are mainly based on the traditional image recognition technology to extract the crack information of digital borehole image,but there are many problems,such as slow extraction speed,easy to be affected by noise,crack does not have continuity,etc.In recent years,with the continuous development of convolutional network,a representative algorithm of deep learning,a new method is provided for image processing technology and a new development direction for crack identification in digital borehole images.In this paper,a crack segmentation model based on the Mask R-CNN algorithm of rock mass drilling image is constructed.Digital drilling images with marked information are used to conduct iterative training on the network model,so that the network model can complete the recognition,classification and segmentation of cracks in digital drilling images.The main research contents are as follows.(1)The digital panoramic drilling imaging system is used to collect a large number of digital drilling images,and the methods of clipping,expansion and noise reduction are used to preprocess the images.labelme software is used to label the cracks of digital drilling images,and the COCO2017 data set of digital drilling images is produced according to relevant formats,which will be used for the training of network models.(2)The basic principle of traditional image processing method is elaborated,and the crack identification of digital borehole image is carried out based on global threshold segmentation,Sobel operator and Canny operator.(3)A digital borehole image crack segmentation model based on Mask R-CNN algorithm is proposed to quickly identify,classify and segment digital borehole image cracks.The network model is trained 500 times iteratively,the loss value generated in the training process is calculated,and the loss curve is drawn.(4)Finally,input the test set into the network model to obtain the prediction result of the test set.Based on the prediction results,the obfuscation matrix,classification evaluation index and PR curve are used to evaluate the network model.Four images were randomly selected,and the recognition results of network model were compared with those of three traditional algorithms,and the advantages and disadvantages of these methods were analyzed.The following conclusions are obtained: When the confusion matrix threshold of the digital drilling image crack segmentation model based on the Mask R-CNN algorithm is 0.7,the recall rate of the network model is 78%,and the accuracy rate is 83%.The two models basically reach a balance,and the F1 score is 80.4%,indicating that the network model has high accuracy and the prediction and classification results of the network model are accurate.The results of the network model are superior to the traditional algorithm in efficiency,accuracy,edge continuity and the degree of influence of non-crack information.Therefore,the network model can be used for the rapid recognition,classification and segmentation of crack information in digital borehole images.