Research On Identification Method Of Rock Fractures In Outcrop Area Based On Deep Learning

Rock cracks are a kind of cracks generated when the rock is subjected to beyond the maximum stress range that it can bear.It can be divided into three types of cracks: "diagenetic cracks","secondary cracks" and "structural cracks".Diagenetic cracks are cracks generated during the process of diagenesis.Secondary cracks are also called weathering cracks.They are cracks generated by the weathering of rocks by temperature,water and biology in nature.Structural cracks are a kind of cracks generated by rocks subjected to crustal movement.Because such fractures can be used to study the evolution of geological structure and predict underground mineral resources in geological exploration,the study of such rock fractures is an important topic in the field of geological research.Fracture information is one of the most basic data in geological exploration.When there is an error in the basic data,it will undoubtedly bring huge deviations to the subsequent model research,causing a lot of waste of physical and human resources,and more seriously,mining The personal safety of industry personnel poses a threat.At present,the acquisition of information on such cracks still mainly stays in the stages of manual hand-painted description and traditional image processing with low accuracy and efficiency.With the development of deep learning in computer image processing,deep learning has been applied to all walks of life.In view of the fact that the information of rock fractures in the field outcrop area is important for geological research,but its identification method is more traditional,which results in poor recognition results,this paper proposes a deep learning-based rock crack identification method in the field outcrop area.The application improves the accuracy and efficiency of rock crack identification.This method is based on Google's Tensorflow deep learning framework.First,the pre-processed training data set images are manually selected and pre-classified into two types of images: cracks and backgrounds,and placed in the corresponding folders of cracks and backgrounds.The pictures in each folder are marked with cracks and back labels and input in batches to the Tensorflow-based convolutional neural network model for training.Secondly,the sliding window algorithm and the standard deviation pre-screening algorithm are used to obtain and pre-screen the preprocessed binary rock crack images.Finally,the trained convolutional neural network model and data are used to identify and classify the unfiltered binary bins obtained by the sliding window algorithm as cracks or backs,display the crack bins and record the position information,omitting the background bins and positions Information,through the recorded binarized crack position information,the crack location of the primary color rock crack image without binarization preprocessing is displayed and displayed.The experimental results show that the accuracy of the method used to identify the rock cracks in the field outcrop area is higher than 91%,which can provide a more accurate and convenient method for crack identification for analytical models such as geological survey and resource prediction.