Resolution is a critical metric for evaluating the reconstruction performance of carbon emission gas concentrations and temperature distributions. Tunable Diode Laser Absorption Tomography (TDLAT) is an effective emission monitoring technique but suffers from limited reconstruction resolution due to sparse detection data. This limitation hinders the acquisition of measurement data in gas flow fields with a high dynamic variability. To address the inherent rank deficiency in sparse tomographic reconstruction, this paper proposes a physics-constraint-guided dual-branch cross-talk UIUNet (PhyDCT-UIUNet). This network enables rapid, high-precision super-resolution reconstruction of CO2 concentration and temperature fields from limited beam data. The network incorporates a cross-attention module to enhance the fusion of shallow-layer details and deep-layer features. It constructs a dual-branch crosstalk module to explicitly model the coupling relationship between the temperature and concentration fields. For the first time, the network integrates the Beer–Lambert law into the loss function to ensure that the reconstruction results comply with physical principles. To evaluate the performance of the proposed PhyDCT-UIUNet, we experimentally validated its reconstruction capabilities across different super-resolution magnifications. Experimental results demonstrate that PhyDCT-UIUNet accurately reconstructs CO2 concentration (temperature) distributions at 4×, 6×, and 8× super-resolution. It achieves average reconstruction errors of 2.65% (2.27%), 4.17% (4.04%), and 5.40% (5.48%), respectively. Compared with previous methods, it exhibits a higher reconstruction accuracy and robust noise resistance. Furthermore, PhyDCT-UIUNet realizes end-to-end mapping from sparse path absorptance data to high-resolution CO2 concentration and temperature fields. This provides a novel solution for highly sensitive, high-precision continuous dynamic monitoring and the imaging of carbon emissions.
Wei et al. (Mon,) studied this question.