Edge Detection Method Of Seismic Data Based On Deep Learning

Seismic structural interpretation is an important section in seismic interpretation,and has been extensively studied in the field of seismic exploration.However,accurate and efficient interpretation of faults and caves is still a very difficult problem and has not been well resolved.With the development of computer technology,the use of computer technology to extract the information of faults and fracture-cavity brings a new perspective.At this stage,the method of automatically detecting faults and caves is mainly by calculating seismic attributes.This type of method provides a basis for the seismic interpretation to a certain extent.However,because this type of method is sentitive to the stratigraphic and noise,the problem of inaccurate detection remains to be solved.In order to improve the effect of the seismic interpretation,this paper introduces a deep learning method.Aiming at the current problems,a method of automatic detection of faults and fractures and caves based on deep learning is proposed.It aims to solve the shortcomings of the existing methods and improve the accuracy and effectiveness of the automatic interpretation of seismic data.The main research content is the identification of faults and fracture-caves.The main research content and innovations of this article are as follows:(1)Apply a set of two-dimensional fault detetcion method based on CNN multi-classification network.This paper divides the faults in the two-dimensional seismic image into seventeen categories based on the dip angle attributes of the faults in the two-dimensional seismic image,and converts the fault detection problem into a fault classification problem.First,cut the two-dimensional seismic section is cut into small seismic data of a certain size,use CNN to predict the attribute of the fault dip in the small piece of seismic image;then use the anisotropic Gaussian function to describe the fault;and finally stack all the locally fault-oriented Gaussian functions to generate a fault proability image,the fault proability image can be obtained.(2)This paper proposes a seismic fault detection based on 3D-Unet3+ convolutional neural network.This neural network is based on the Unet convolutional neural network as a prototype.It is improved to a 3D-Unet3+ through the use of full-scale connected and 2D to3 D upgrades.By referring to the advantages and disadvantages of the 3D-Unet fault detection method,this paper uses field seismic image to make a fault label.Different from the manual method of marking fault,it uses automated marking fault,which not only maintains the accuracy of marking,but also improves the efficiency of marking the fault.In order to avoid the problem of the unequal number of faults and non-faults in the fault label,a class-balanced binary cross-entropy loss function is used to improve the accuracy of fault detection.Finally,the better performance of the method in field seismic image verifies the effectiveness and accuracy of the method.(3)This paper proposes a seismic fracture-cavity detection based on 3D-Unet3+convolutional neural network.This method use the same network structure as the fault detection method based on 3D-Unet3+.By using synthetic seismic images and the corresponding label images of the collapsed paleokarst features simulated in the seismic images as the training set method,the network structure is trained,and then the field seismic image is predicted.Finally,through the application of this method,the field seismic image in western my country is detected,and the detection result is better than the traditional seismic attribute method.At the same time,the detection method based on the3D-Unet3+ network is very efficient,and it only takes a few minutes to detect a226*382*600 three-dimensional data volume.