GLRNet: Gas Leak Recognition via Temporal Difference in Infrared Video

Erqi Huang, Linsen Chen, Tao Lv, Xun Cao

Abstract: Gas Leak Recognition (GLR) is a task where existing techniques face significant challenges, such as the need for infrared cameras, various industrial scenes, and validated data. In this work, We demonstrate Gas Leak Recognition Network (GLRNet), a network with temporal difference inputs and temporal shifting operations. GLRNet integrating module is designed by analogy to human perception, which is a physical constraint in the feature representation. The synergy of our proposed GLRNet and infrared camera we have developed is an emerging comprehensive system that achieves state-of-the-art (SOTA) results in real-world data captured in several chemical industrial parks across the country. Our demo video is at https://youtu.be/glt4DMeNXDU .