84th EAGE Annual Conference & Exhibition
Summary
This research introduces a new approach, Image-to-image translation (I2I) using GAN as the foundation to solve the problem of missing traces in seismic data caused by shallow gas or other factors. I2I uses a zero-sum game between a generator model and a discriminator model. The workflow begins with conditioning and preparation of input and output datasets. The data is then normalized and used to train the model for 1500 epochs or 25 minutes using a GPU. The model is able to predict missing traces in seismic data with high accuracy, preserving the fault delineation and excluding migration artifacts present in the original data. The results are further validated by comparing the amplitude spectrum and cross-correlation coefficient between the original data and the predicted missing trace. The predicted missing trace is highly preserved with cross-correlation value of over 0.8 depending on the trace location. This approach is an advancement over traditional interpolation techniques and can help geoscientists to better understand the structures below the surface.