Gas Leak Real-time Detection and Volume Flow Quantification Based on Infrared Imaging and Advanced Algorithms

• Published in: IEEE access 2025-01, Vol.13, p.1-1
• Main Authors: Yan, Man, Li, Zhou, Dong, Zheng, Liu, Yiming, Chen, Liyun, Wu, Xiaosong, Wu, Lijun
• Format: Article
• Language: English
• Subjects: Accuracy; Cameras; Clustering algorithms; DeeplabV3+ neural network; Fluid flow; Gas identification; Gas lasers; Gas quantification; Gases; Image edge detection; Kmeans clustering algorithm; Optical filters; Optical flow method; Real-time systems; Training
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2025.3525764

Due to the semi-transparent and irregular nature of gases, it is still a highly challenging task to effectively detect and quantify gas leaks especially those with small flow rates by only utilizing economical equipments. In this paper, we present a strategy for automating real-time identification and quantification of gases in the mid-infrared band by combining an infrared camera combined with a series optimized algorithms. A basic network DeepLabV3+ is first modified by replacing its Xception backbone with MobileNetv2 for real-time gas detection and segmentation. Then special attention mechanisms tailored to the characteristics of the gas are added into the network to enhance the perception and recognition of the gas edges. The optimized Kmeans clustering algorithm is integrated to identify the Region of Interest (ROI) in the image containing the target gas. The quantification of the volume flow rate within the ROI is realized by integrating the radiation transfer model with the optical flow method. The experimental results indicate that the quantification limit of the gas flow rate can reach 0.01 L/min, which is comparable to that obtained by the methods with complicated instruments. Our detection and quantification strategy can find vast applications in hazardous gas monitoring field.