Optical Gas Imaging Detects Leaks Using Channel Stacking

_

This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 229227, “Enhancing Optical-Gas-Imaging Leak Detection by Transforming Temporal Dynamics Into Spatial Features Using Channel Stacking,” by Mehdi Korjani and Jaeyoon Chung, Clean Connect. The paper has not been peer-reviewed.

_

While optical gas imaging (OGI) is indispensable for visualizing subtle and diffuse invisible hydrocarbon gas plumes, accurately identifying leaks from single, static infrared frames is often compromised by the inherent limitations of single-channel data and the complexity of industrial backgrounds. Full video processing can capture essential plume motion; however, its significant computational overhead typically prohibits real-time deployment and misinterprets non-gas-related movements as a gas leak within the scene. This paper introduces “channel stacking,” a novel, efficient methodology that transforms critical temporal gas‑motion dynamics into robust spatial features.

OGI in Emissions Detection

OGI offers a noninvasive means of visualizing fugitive gas emissions invisible to the naked eye. However, traditional OGI surveys rely heavily on manual interpretation by trained technicians. This process is inherently subjective, labor-intensive, and time-consuming. Studies have shown that highly experienced surveyors can detect up to 1.7 times more leaks than their less-experienced counterparts, highlighting a critical vulnerability in relying on human factors for regulatory compliance and environmental protection.

This issue has catalyzed a significant research and development effort toward automated OGI‑ analysis systems powered by artificial intelligence and deep learning. However, automated analysis of OGI data presents a tradeoff. Analyzing single, static frames is computationally cheap but often fails to capture the subtle motion that distinguishes a gas leak from background noise. Conversely, processing full video with temporal deep-learning models can capture this motion, but is too computationally expensive for real-time edge deployment, and can misinterpret non-gas movements. By transforming the temporal dynamics of a gas plume into a static, multichannel spatial representation, the authors’ channel-stacking approach allows a deep-learning model to detect motion without the overhead of video processing. Rigorous analysis has demonstrated that this framework not only automates OGI analysis with high accuracy, but also exhibits exceptional robustness against the diverse environmental and operational challenges that have plagued automated detection systems.

Model Development

Data.

The model was trained using a comprehensive, proprietary data set that builds upon and significantly expands the more than 10,000 videos used in previous studies. This new data set encompasses an even wider spectrum of real-world variables, including diverse industrial backgrounds, a full range of leak rates from small fugitive emissions to large-volume releases, and a vast array of environmental conditions (clear, rain, snow, fog, varying wind speeds, and day or night cycles).

To further enhance the model’s resilience, a suite of advanced data-augmentation techniques specific to the OGI domain was employed during training. These techniques include simulating variations in thermal contrast, modeling different plume‑dispersion patterns based on wind conditions, and realistically blending plume signatures into complex backgrounds. This rigorous training regimen ensures that the model generalizes well and is prepared for the challenges of field deployment.