SM Sai Kiran Maryada
Research Scientist · Data Institute for Societal Challenges

Publications

Recent publications and conference contributions. The full record is on Google Scholar.

Nov 2025Conference

Deep Learning-Based Estimation of Petrophysical Properties for Unconventional Shales from SEM Images

ADIPEC — Abu Dhabi International Petroleum Exhibition and Conference
L. Livingstone, P. Yeturi, I. Cherukuri, D. Devegowda, M. Curtis, C. Rai, S.K. Maryada

A deep learning workflow that rapidly estimates petrophysical properties in unconventional shales directly from SEM images. Using 8,000+ images from 22 shale plays, a ResNet50-based classifier learns microstructural differences; its feature embeddings are then related to porosity, mineralogy, and elastic properties — avoiding segmentation workflows that can take weeks per image.

Jun 2025Journal · Q1

An Improved Data-Driven Method for the Prediction of Elastic Properties in Unconventional Shales from SEM Images

Geoenergy Science and Engineering, Article 214043
S.K. Maryada, D. Devegowda, C. Rai, M. Curtis, D. Ebert, G. Danala

A fast, image-based workflow to estimate Young’s modulus from SEM images of drill cuttings across 14 shale plays. Textural and shape-based features are combined with deep learning representations and correlated with measured elastic properties via non-parametric regression, yielding strong predictive capability that holds on previously unseen images.

Jun 2025Conference

Self-Supervised Learning Using Vision Transformer Architecture for Rock Image Segmentation

SPE Europec @ EAGE Conference and Exhibition
R.D. Mohammad, D. Devegowda, C. Rai, M. Curtis, S. Mudduluru, S.K. Maryada

A self-supervised ViT-based segmentation workflow for FIB-SEM shale images that removes the need for pixel-level labels. The model segments images into 12 sub-classes grouped into organic, inorganic, and pore space, performing robustly across 22 shale plays and significantly accelerating shale characterization.