Research On Geological Anomaly Identification Method Based On Deep Learning

Geological anomalies usually refer to geological bodies that cause geophysical detection anomaly.They are special geological structures with a certain spatial form.Salt domes and channels are the most common geological anomalies.Due to the special structure of geological anomalies and the close connection with resources such as oil and gas reservoirs,the identification and segmentation of geological anomalies have a crucial role in seismic interpretation.The 3D model of geological anomaly body is of great significance for seismic migration velocity model,mineral resources prediction and geological structure analysis.At present,a lot of identification of geologic abnormal body was conducted based on a single attribute.Although there are multi-attribute fusion methods,they all use algorithms to do multi-attribute fusion,which depends on manpower and takes a lot of time,it is difficult to meet the current demand of exploration and development.On the other hand,with the development of artificial intelligence,the research of identifying geological anomalies based on convolutional neural networks has gradually become a research hotspot,but the current research is all based on a single attribute.However,the single attribute cannot comprehensively and accurately reflect the geology structure and details of the anomaly and has multiple solutions.Therefore,based on seismic attributes,this thesis explores the use of deep learning methods for multi-attribute fusion to achieve more automatic and accurate segmentation of geological anomalies.This thesis studies salt domes identification and segmentation based on 3D seismic data.The main work and contributions are as follows:(1)Based on the low efficiency of manual interpretation and the multiple solutions of single attribute identification,this thesis proposes a salt domes identification method based on Unet and multi-attributes fusion.Firstly,several attributes are selected according to the sensitivity to the geological anomalies to be identified,and preprocessing operations are performed according to the characteristics of the seismic data.Later,by constructing a Unet network with multiple inputs and multiple connections,attributes data can be entered into the network at the same time and fused in the network.The network implements end-to-end salt domes segmentation,and considers the non-linear problem of multi-attribute fusion using dense connections.Unet using multi-attribute fusion can automatically fuse and learn multi-attribute features,and the accuracy of salt domes identification is better than Unet with single attribute.(2)It is necessary to construct three-dimensional geological anomalies in seismic interpretation.In order to solve the problem of non-smoothness when constructing a three-dimensional model from two-dimensional segmentation results,this thesis further proposes a salt domes identification method based on 3DUnet and multi-attributes fusion and establish the overall workflow for the construction of 3D anomaly models.First of all,in order to solve the problem of difficulty in labeling 3D data,the minimum curvature algorithm is used to smooth the results of the 2D salt segmentation as labels for the 3D salt mound segmentation.Second,the salt recognition method based on Unet and multiattribute fusion was extended to 3D.All 2D operations are replaced with 3D,which realized the direct segmentation of 3D data.In order to enhance the accuracy of the recognition results,a region-based growth algorithm is proposed to further optimize the segmentation results.Finally,the isosurface extraction technique is used to reconstruct the three-dimensional salt mound surface.This chapter realizes the multi-attribute fusion of 3D seismic data,obtains a smooth and accurate 3D segmentation result,and realizes the workflow of 3D salt identification and reconstruction..The two-dimensional operation was replaced with three-dimensional operation,which realized the direct segmentation of three-dimensional data.At the same time,in order to solve the problem of difficult 3D data labeling,the minimum curvature algorithm is used to smooth the 2D salt mound segmentation results as labels for the 3D salt mound segmentation.In order to enhance the accuracy of the recognition results,a region-based growth algorithm is used to further optimize the segmentation results.This chapter realizes the multi-attribute fusion of 3D seismic data and obtains a smooth and accurate 3D segmentation result.In the interpretation of seismic data,it is very necessary to construct a threedimensional anomaly model.In order to solve the problem of unevenness when constructing a three-dimensional model from the two-dimensional segmentation results,a salt dune recognition method based on 3DUnet and multi-attribute fusion is proposedIn this thesis,the proposed method is verified by Netherlands off-shore F3 block seismic data.The salt domes segmentation results obtained are basically consistent with manual interpretation results.Compared with the Unet network using a single attribute,the accuracy of the segmentation results obtained by the proposed method is higher,and the boundary of the geological anomalies obtained by the proposed method is clearer.