Research On Crack Identification And Quantitative Statistics Of Rock Mass Based On Deep Learning

Cracks in rock mass can change the stress state of rock mass,resulting in deformation,destruction and instability of rock mass.Crack damage is a research hotspot in engineering geology and mining engineering.Crack detection and statistics are the basis for studying rock mass crack damage.Traditional crack detection methods have high false detection rate,low efficiency and poor effect.Crack detection and range quantification has always been a difficult problem in the industry.Research on computer methods of rapid crack detection to achieve on-line crack detection.It has certain theoretical value and practical significance for the construction of intelligent mines.For the problem of computer method for fast crack detection,YOLO X algorithm for different crack location detection is constructed by deep learning method,U-Net crack image boundary segmentation network model is established,U-Net algorithm is improved,quantitative statistics of crack area is realized,and the algorithm is verified by using crack image of surrounding rock of mine roadway in Gongchangling underground mine,and the following conclusions are obtained:(1)Through the research of target detection algorithm of crack image,the number of rock mass crack images is expanded by combining traditional method with sliding window algorithm,which solves the problem of too small capacity of image data collection of cracks.Crack types are divided according to different crack shapes.By building YOLO X algorithm model,different types of crack images and videos are located and detected.Although this algorithm can accurately identify the fracture type,it is difficult to obtain the fracture texture shape and realize the quantitative statistics of the fracture range.(2)Based on the semantic segmentation algorithm and its improvement,the U-Net image segmentation algorithm is constructed.By comparing the crack detection effect of VGG16 and ResNet50 as the backbone feature extraction network,a rock mass crack identification model with higher segmentation accuracy and better detection effect is obtained.More than 85% of the mIoU evaluation indexes were obtained twice on the data verification set.The model was used to detect the images and videos with different morphological cracks,achieving a good segmentation effect.(3)Through the study of quantitative statistical method of rock mass cracks,a quantitative statistical method of crack area based on projection mapping is proposed,which achieves the division and area quantification of different types of cracks.Through the application research of roadway rock mass crack recognition algorithm,the improved U-Net algorithm model is used to detect the surrounding rock cracks in the iron mine roadway of Gongchangling underground mine,and the area of local rock mass cracks is quantitatively counted.Compared with the traditional image processing algorithm and FCN,DeepLabV3+ algorithm,the results show that the improved UNet algorithm model is more accurate and has better recognition effect on the crack boundary.