Vision-Based Analytics of Flare Stacks Using Deep Learning Detection

Flare stacks play a critical role in oil refineries and chemical plants, but monitoring their performance is a challenging task that often requires skilled operators. To address this challenge, we propose a novel approach that combines video capturing and machine learning techniques to automate the monitoring of flare stack operations in real-time. Our vision-based system analyzes captured video footage of the flare stack's scene and employs state-of-the-art deep learning detection models, including YOLOv5, YOLOv7, and the Detection Transformer (DETR), to detect and analyze combustion-related objects such as flame and smoke. Rigorous experiments show that the proposed technique was able to accurately detect flame and smoke objects in flare stacks scene and the best model showed encouraging performance metrics. By leveraging the power of recent deep detection models, our proposed system offers a promising alternative to labor-intensive manual inspection by keeping a continuous and automated watchable eye in combustion quality, facilitating more efficient and reliable flare stack operation analysis.