This project using Netherlands Offshore F3 Block as the real seismic field dataset to apply the CNN model, link to dataset. We applied CNN model to predict 10 seismic facies using 3D seismic data as the input. Then preprocessed the seismic data by normalize the data and divided into small cube to feed into CNN model. After got the prepreocessed dataset, we split the data for training, testing, and validating processes.
We train CNN model using U-Net architecture with transfer learning method. After tuned the hyperparameter including transfer learning pretrained model, optimizer, activation, and batch size, we achieved 88% of accuracy for testing dataset. This model reached the optimum accuracy after training using 150 epochs or 33 minutes with the GPU.
Skills: Python, Numpy, Pandas, scikit-learn, Keras, Scipy