Facies classification using semi-supervised deep learning with pseudo-labeling strategy

• DOI: 10.1190/segam2019-3216086.1
• Publication Date: 2019-08-10T19:18:37Z
• Authors: Asghar Saleem; Junhwan Choi; Daeung Yoon; Joongmoo Byun
• Keywords: Labeled data; Supervised Learning; Lithology

Quantitative facies classification is the key to linking seismic data lithology evaluate important reservoir properties. During past several years, size of volumes has piled up extent that it challenging for experts examine every volume classify facies. This motivated machine learning approach predicting in an efficient way. However, labeled (well data) limited by various constraints and very expensive obtain, whereas, there a plethora unlabeled (seismic data). Geophysicists are tasked interpret enormous amount unclassified on basis sparse data. In this study, we have adopted Semi-Supervised Learning using pseudo-labeling analysis order overcome scarcity leveraging With each step, small classified starting from vicinity well gradually moving away well. The (called 'pseudo label data') added used retraining classifier, adding diversity classifier accounts lateral change while Following our proposed workflow, shown accuracy trained can be enhanced considerably combining number with large pool inverted technique. Furthermore, results applying workflow field data, outperforming conventional methods, achieved 99.69% loss as low 0.001 both training validation, moreover, task carried out error 0.004. Presentation Date: Monday, September 16, 2019 Session Start Time: 1:50 PM 2:40 Location: 217A Type: Oral