From Pixels to Insights- Harnessing the Power of Deep Learning for Facies Classification in Borehole Images

Abstract

Abstract The objective of this study is to develop a deep-learning-based framework for automated facies classification in borehole images. We aim to address the limitations of manual facies classification, including subjectivity, errors, and time consumption. The framework should be efficient, accurate, and easily integrable into existing workflows in the oil and gas industry. The study uses deep-learning techniques, specifically convolutional neural networks (CNN), to develop a model for feature extraction and classification. The model employed is the state-of-the-art, open-source EfficientNetB4 architecture. The study also employs data preparation techniques, such as data augmentation and Synthetic Minority Over-sampling Technique (SMOTE) to handle labelled data set class imbalance. The study involves three main steps: data preparation, model training, and evaluation. In the data preparation step, borehole images are processed into pixel-level data, sampled at a particular resolution, labelled, and augmented. In the model training step, the model is trained on the labelled data set using the EfficientNetB4 architecture. The evaluation step involves measuring the accuracy and F1 score of the model to analyze its performance. The results demonstrate the effectiveness of the proposed framework, achieving an accuracy of 85% on the test data within minutes. The study also shows that the model can generalize well to variations in image quality and resolution, indicating its robustness and potential applicability in real-world scenarios. The study highlights the potential of deep learning for automated facies classification in borehole images, providing a more objective, accurate, and efficient approach to subsurface characterization. The use of artificial intelligence (AI) and analytic platforms enables efficient integration of the framework into existing workflows, supporting decision making in the oil and gas industry. The study demonstrates that deep-learning-based frameworks can offer valuable insights for the oil and gas industry, enabling improved resource management and accelerating project delivery. The proposed framework provides an objective and efficient approach to facies classification, overcoming the limitations of manual techniques. The study also shows the potential for the framework to be applied to other image classification tasks in the oil and gas industry.