Surface Detection of Oil and Water from Underground Pipelines

This project, in collaboration with HPCL, IOCL, and GAIL, focuses on detecting surface signatures of oil and water leaks originating from underground pipelines using multispectral imaging and AI-based analysis.

Problem Statement: Pilferage Detection

A significant challenge faced by oil and gas companies is the undetected pilferage and leakage of oil from underground pipelines. Early surface-level detection can prevent environmental damage, operational downtime, and economic losses.

Preliminary Spectral Analysis

Initial analysis focused on identifying spectral bands that effectively distinguish between oil and water on soil and vegetation. Multispectral imaging helped isolate key wavelengths where oil and water show distinct reflectance patterns.

Dataset Creation and Model Development

A custom multispectral dataset was created under controlled conditions using varying quantities of oil, water, and vegetation. This comprehensive dataset enabled the training of a robust deep learning model based on the YOLOv8 architecture. The model achieved an accuracy of 92.9% for oil detection and 96% for water detection, showcasing strong performance in surface leak classification tasks.

Phase 2: Real-World Deployment

In the second phase of the project, we plan to augment the current dataset with real-world data collected from actual pipeline leakage sites, in collaboration with industry partners such as IOCL, HPCL, and GAIL. This phase aims to validate and improve the model's performance under diverse environmental and terrain conditions. Field data will include site-specific spectral signatures, soil types, and contamination patterns to ensure the model generalizes well to operational settings.

In parallel, comparative benchmarking against other state-of-the-art deep learning algorithms is ongoing, and the results will contribute to a scientific publication currently under development.