| Abstract Submission No. | ABS-4.1-0097 |
| Title of Abstract | Subsurface Fault Interpretation Using Convolutional Networks in the Upper Assam Foreland Basin, Northeast India |
| Authors | Dr. Jitender Kumar*, Dr. Kalachand Sain |
| Organisation | Wadia Institute of Himalayan Geology, Dehradun |
| Address | Wadia Institute of Himalayan Geology DEHRADUN, Uttarakhand, India Pincode: 248001 E-mail: jitender@wihg.res.in |
| Country | India |
| Presentation | Oral |
| Abstract | Understanding subsurface faults is vital for studying reservoir heterogeneity, locating hydrocarbons, tracking basin changes, maintaining reservoir stability, and evaluating seismic risks. Traditional seismic interpretation workflows can be time-consuming and may be influenced by interpreter bias. In this study, we use high-quality 3D seismic reflection data from the Upper Assam foreland basin to automatically find and map subsurface faults and fractures. We developed a machine learning framework trained on real, labeled seismic data. The convolutional neural network (CNN) learns spatial features that show fault shapes, such as breaks, angles, and ends. The objective is to enhance the accuracy and efficiency of fault detection, minimize interpreter bias, and promote more consistent and scalable subsurface fault modeling. In the present study, utilizing the fault mask by Edge preserving smoother filter (EPS) and thin faults likelihood (TFL) to demonstrate the fault networks, outperforming traditional edge detection and attribute-based methods in both accuracy and speed. The fault networks divided the Oligocene-Miocene period into several structural sections. Thus, the method allows for fast, consistent, and scalable fault detection, which could improve automated seismic interpretation in geoscience and petroleum work. |
| Keywords | EPS, TFL, CNN, Upper Assam Basin |