Use of Image-Based Machine Learning and Quantum Learning for Subsurface Characterization

Abstract

The proposed thesis aims to explore novel applications of machine learning for subsurface characterization. In the first chapter, an image-based data-driven workflow is proposed to characterize oil viscosity from side-wall rock sample images. Informative features are extracted from the rock sample images deploying several image-based filters and statistical models. Both regression and classification tasks are performed on the preprocessed data. The proposed workflow shows promising results for viscosity classification whereas future work is needed to improve the regression performance. The second and third chapters explore the application of quantum-enhanced machine learning models for lithology classification and the resulting comparison with classical machine learning models. The second chapter compares a quantum support vector machine with a traditional support vector classifier for lithology classification from well log data. Different sample sizes are tested to understand if a quantum advantage is obtained when the available data is limited. The third chapter investigates the application of both quantum support vector and variational quantum classifier for binary lithology classification. The score distribution obtained from testing the models with multiple iterations gives more insight on the current performance capabilities of quantum-enhanced machine models when compared to artificial networks. Overall, although a quantum advantage is not observed in both chapters, this work opens the door to future applications of quantum-enhanced machine learning for subsurface characterization.