Salt Dome Identification And Seismic Facies Classification With Deep Learning

Salt domes identification and seismic facies classification are of great significance to oil and gas exploration.The salt dome is an important oil storage structure,and a large amount of oil is stored in the inner core.The classification of seismic facies helps to better analyze the geological stratigraphic structure and improve the reliability of oil and gas exploration.Traditional methods for the study of geological bodies combine seismic attributes with researchers 'theoretical knowledge and experience.This method is inefficient and inaccurate.With the development of deep learning,the characteristics of self-learning and self-adaption of neural networks have effectively promoted the study of geological bodies and improved the speed and quality of oil and gas exploration.Because the geological volume data collection is noisy and sensitive to reflection,as well as the neural network has strict requirements for data labeling,it is difficult to obtain good results by directly applying the deep learning model.In response to these problems,this paper studies the identification of salt domes and the division of seismic facies.The main work includes:(1)Salt domes recognition and segmentation based on weakly supervised learningIn view of the lack of labeling of seismic data,this paper has designed a set of salt domes recognition and segmentation processes that apply weakly supervised learning with bounding box labeling.The algorithm flow includes three stages: pre-processing,model training and post-processing.The preprocessing calculation enhances the supervised information of the salt domes data.The recursive training method of the model reduces the distance from the performance difference of the fully supervised learning.Post-processing strictly controls the model prediction results.The entire process model self-learns the characteristics of the salt dome and recognizes the outline and shape of the salt dome without manual intervention.The algorithm verified the effectiveness of this process on the data set provided by the TGS Salt Identification Challenge competition on the kaggle platform,and the result can reach 92.57% under full supervision.(2)Recognition and division of seismic facies based on improved Unet modelAiming at the problem that the Unet model crosses the connection between the shallow layer and the deep layer and causes the middle network layer to be bypassed,this paper proposes a seismic phase division method based on the improved Unet model.Based on the original Unet model,the structure and parameters of the seismic data are adjusted,so that the network model can extract the seismic facies features in more detail.The added cross-layer connection expands the receptive field of neurons in the network,and solves the problem that the Unet model is insensitive to fine-grained features.This algorithm has been verified on the feasibility of the North Sea Dutch dataset F3.The performance of the model has been greatly improved,and the accuracy of seismic facies classification has also been improved.