Detection of Methane Leaks by the Use of Mid-range Infrared Camera

Timely detection of methane leaks from natural gas infrastructure, like pipelines and valves, is essential for mitigating environmental and safety risks. This research focuses on using mid-wave infrared (MwIR) cameras on unmanned aerial systems (UAS) to detect leaks. By leveraging machine learning, the study aims to develop an efficient, real-time methane inspection system that operates directly on embedded processors on UAS platforms. This project employs the FLIR G300a OGI camera for remote gas inspection, utilizing video preprocessing and optical flow to mask gas plumes in footage. The YOLOv8 deep learning model is used to detect gas pixels and segment the plume area, with the analytics deployed onboard an NVIDIA Jetson Nano. Indoor experiments demonstrated that the FLIR G300a camera can effectively detect low methane flow rates (3-5 SCFH) at a 30 ft distance, highlighting its strong performance under controlled conditions. In outdoor settings, detection proved more challenging due to factors like wind, temperature, and complex backgrounds. However, preprocessing videos and applying optical flow significantly improved gas pixel identification, enhancing labeling and training for the YOLOv8 model, and demonstrating the system’s adaptability to real-world environments. The developed methodology automates the previously labor-intensive and hazardous task of detecting methane leaks with handheld sensors, which require close proximity to the leak source. This remote, automated approach enhances efficiency, enabling fast response and timely mitigation of environmental, safety, and property risks. Additionally, the technology can be adapted to detect other gasses around chemical and industrial facilities.