A research team from the Hefei Institutes of Physical Science of the Chinese Academy of Sciences has developed a fast, multiplatform-compatible detection network that can "see" gas leaks in three dimensions.

The results were published in Environment International and Remote Sensing.

Frequent fires and explosions caused by gas leaks have drawn increasing attention to detection and monitoring. However, existing gas remote sensing systems are usually limited to two-dimensional (2D) projections and are unable to quickly provide essential three-dimensional (3D) information, such as gas volume, distribution, diffusion, and source location.

To address these limitations, the researchers focused on gas species identification, leak localization, and accurate volume quantification. For rapid leakage scenarios, they built a multispectral imaging system integrating infrared detectors, lenses, and motorized components. Using the YOLOv10 model, the system achieves real-time detection at over 25 frames per second. 

Combined with a non-axisymmetric inverse Abel transform, 3D reconstruction was completed within 200 milliseconds. Simulations demonstrate high reconstruction accuracy with a peak signal-to-noise ratio (PSNR) of 25.633 and a structural similarity (SSIM) of 0.940.

For large-scale gas plumes, the team developed the ZK-FTIR-GS1000 imager and a deep learning-based 3D reconstruction network. Using an octree representation, the network reconstructs gas clouds with minimal computational resources, going from coarse to fine resolution. Field tests demonstrate that the network effectively captures the spatial location and distribution of gas leaks, rendering it suitable for large-scale applications.

This work provides robust tools for the rapid 3D detection and spatial reconstruction of gas leaks, offering strong technical support for environmental monitoring, emergency response, and industrial safety.

Schematic diagram of the multispectral imaging system (Image by XU Liang)