Research On Analysis And Recognition Method Of Stratigraphic Detection Image Based On Deep Learning

There are many ways to visit underground structures.The cracks and fracture zones on the surface can be directly observed and quantitatively measured and described by various equipment.As for the structure in the rock mass,it must be observed with the help of specific technical means,and the structure in the rock mass is the key factor to determine the structural characteristics of rock strata,which is of great significance to geological engineering research.Forward-looking borehole camera system can acquire borehole images and videos in rock mass,which is an important means to observe the internal structure of rock mass.How to use these borehole images and video data to directly identify cracks is an important research topic.In this paper,a mapping method of forward-looking borehole images is designed,and deep learning technology is used to identify cracks in rock mass.The main research contents are as follows:(1)In order to meet the data needs of forward-looking borehole expansion and splicing,the research compares the forward-looking camera technology and the digital panoramic optical camera technology,and uses the forward-looking camera technology to obtain the drilling data.(2)In order to qualitatively and quantitatively analyze the hole wall images and eliminate the distortion caused by forward-looking drilling images,a drilling expansion method that adaptively determines the drilling center and drilling expansion radius is proposed.Clarify the original borehole image by using CLAHE filtering method;the adaptive positioning method of drilling center is put forward,which achieves the effect of fast positioning of drilling center;the method of circle unfolding is improved,and the borehole image can be unfolded automatically and quickly.Effectively eliminate the influence of camera shake and rotation.(3)In order to observe the hole wall in the image more conveniently and comprehensively,and obtain the expanded image inside the hole wall with a wide range of viewing angles,comparing the template-based matching stitching method with the feature-based matching stitching method,an automatic matching and splicing method based on gray value for the development diagram of forward-looking borehole wall is proposed,by using the gray difference of the image combined with the weighted mosaic fusion method,the forward-looking borehole development map can be quickly and accurately spliced.(4)In order to speed up the identification efficiency and detection accuracy,this paper proposes a method for identifying cracks in forward-looking borehole wall based on deep learning model.Realize the automatic detection of cracks in borehole wall images.The experimental results show that the detection accuracy of up to 92.46% can be obtained by using this method,which is of great significance in actual geologic engineering.There are 52 drawings,5 tables and 72 references.