Spatial location and distribution reconstruction of the leaking gas plume via a single infrared remote sensing system

Spatial localization and distribution of leaked gases are critical for environmental monitoring and emergency management. Infrared remote sensing is widely used for gas leak detection because of its unique advantages. However, a single infrared remote sensing system only measures the two-dimensional projection of gas concentration. Most methods for spatial localization and distribution reconstruction of gas plumes require data from multiple instruments or multi-angle measurements, which increases deployment costs and complicates reconstruction processes in practice. Additionally, the resolution of infrared remote sensing instruments and computer storage capacity limit the achievable spatial resolution of gas plume reconstructions. To address these challenges, this paper proposes a deep learning-based generative network for three-dimensional gas plume reconstruction. The network employs an octree representation to model the sparse three-dimensional distribution of gas plumes. It generates outputs from coarse to fine scales while requiring minimal computational and memory resources. The network takes measurements from a single remote sensing system and predicts a finer octree structure of the gas plume. Field experiments demonstrate that this method can determine the spatial location and distribution of leaked gas plumes, providing effective support for air pollution control efforts.