Research And Implementation Of Fault Recognition Method Based On Image Segmentation

The identification of faults has always been the focus of seismic exploration.With the growth of oil demand,the recognition method of faults has been continuously improved from manual interpretation to automatic attribute interpretation,and then to intelligent deep learning interpretation.In order to further improve the recognition effect of faults,this article based on the idea of image segmentation,using deep learning technology to realize the automatic intelligent recognition of faults.Focusing on the small proportion of fault information in seismic data,on the one hand,the fault information can be learned more fully by fusing multiple different geological attributes,and on the other hand,it can reduce the loss of fault information during the network training process.In this article,two different convolutional neural network models are used to identify seismic data with two different interpretation modes,two-dimensional and threedimensional.The specific work carried out is as follows:(1)Aiming at the information contained in seismic amplitude attributes is limited,and a single network cannot effectively learn the characteristics of the fault zone in seismic data,this article proposes a two-dimensional fault identification method based on multi-attribute fusion.Based on U-Net,the MA-Unet model is built.MA-Unet can perform fusion learning of multiple attributes in seismic data.Aiming at the imbalance in the proportions of faulty and non-faulted samples in seismic data,the Lovasz loss function is introduced to improve the accuracy of fault recognition.Aiming at the problem that the convolutional local receptive field will lose the global information,a Non-local component is added to the network to learn the global information of the seismic data.Finally,the experimental results of the algorithm in the actual work area data verify its effectiveness and practicability in the field of fault recognition.(2)Aiming at the loss of fault information during the training of convolutional neural networks,this article proposes a D-Vnet three-dimensional fault recognition method based on the V-Net network.The algorithm improves the V-Net model,replaces the original convolutional layer in the residual module with an expanded convolutional layer,and builds an expanded residual convolution module,which solves the problem of reduced resolution of seismic data during training.At the same time,according to the specific characteristics of the geological data,the network model is simplified,and the instance normalization layer is added,which further improves the accuracy of fault identification.In addition,the loss function of the network is improved to the Lovasz loss function,which directly optimizes the measurement index Io U of image segmentation.Finally,through theoretical data training and actual work area data prediction methods,the effectiveness and practical value of this method are proved.