Geological Features Estimation with Semi-Supervised Learning

This project is a part of my master’s thesis. The main problem with classification problems is gathering the right amount of labels or ground truth. Meanwhile, in the geoscience world, labeling the geological features requires time and expertise, which can lead to biases in the model.

This project can gather the right amount of labeled datasets by combining the seismic multiattribute to enlighten the targeted geological features and unsupervised learning methods, including PCA and k-means clustering.

Next, this labeled dataset, together with the original seismic, will be used as output and input to build the Convolutional Neural Network (CNN) with U-Net architecture to detect those geological features in 3D seismic data automatically. The result will be evaluated with the seismic geomorphology analysis and finds out that this result gives us the right and consistent interpretation with the manual interpretation.

Skills: Python, Numpy, Pandas, scikit-learn, Keras, Scipy