Hydrogen leakage diagnosis using background-oriented schlieren with dual-channel MGR-Net and transfer learning

Accurate hydrogen leakage detection is critical for the safety of high-pressure systems. Background-Oriented Schlieren (BOS) is a powerful visualisation tool, but its quantitative accuracy, particularly the precision of displacement field extraction, is severely limited by the viewing distance, compromising reliability in far-field measurements. To overcome this fundamental limitation, this study proposes a novel dual-channel MGR-Net framework that synergistically integrates near-field and far-field data. By incorporating transfer learning to accelerate convergence and enhance feature extraction, and employing D-S evidence theory for decision-level fusion of the dual-channel outputs, our method achieves a state-of-the-art diagnostic accuracy of 98.94 %. This represents a significant improvement over traditional BOS techniques, offering a robust and intelligent solution for precise hydrogen leakage identification.