Keywords
Non-flat surfaces
Deep learning
Emissivity correction
U-Net
Abstract
Infrared thermal imaging of non-flat surfaces presents a significant challenge in the field of thermography. Conventional methods suffer from substantial errors when applied to complex geometries such as concave surfaces and steps, primarily due to the spatial variability and viewing-angle dependence of surface emissivity. Building upon the 4D thermal imaging system proposed by Huang, Yang, and Zhu (2023)—which integrates a binocular structured-light camera with an infrared thermal camera—this study introduces a convolutional neural network (CNN)-based method for emissivity correction and thermal image reconstruction. A simulation dataset encompassing a variety of non-flat geometric features is constructed to train a U-Net architecture for pixel-level emissivity correction, which fuses corrected thermal data with three-dimensional surface geometry information to produce reconstructed thermal images. Simulation results demonstrate that the proposed method reduces temperature measurement error by approximately 23.5% on average compared to the conventional geometric calibration approach, with particularly pronounced improvements in concave corners and shadowed regions. This work offers a post-processing solution for non-flat surface infrared thermography that requires no additional hardware modification.
References
1. Apogee Instruments. (2024). *A review of the physics for emissivity correction of infrared temperature measurements*. Ap-ogee Instruments. https://www.apogeeinstruments.com/
2. Chen, Z., Zhang, M., & Li, W. (2020). Advances in industrial inspection applications of infrared thermal imaging technology. *Infrared Technology*, 42(3), 215–223. (In Chinese)
3. Duan, Y., Liu, Z., & Li, M. (2023). Super-resolution reconstruction of sea surface pollutant diffusion images based on deep learning models: A case study of thermal discharge from a coastal power plant. *Frontiers in Marine Science*, 10, 1211981. https://doi.org/10.3389/fmars.2023.1211981
4. Huang, H., Yang, Y., & Zhu, Y. (2023). Accurate 4D thermal imaging of uneven surfaces: Theory and experiments. *International Journal of Heat and Mass Transfer*, 216, 124580. https://doi.org/10.1016/j.ijheatmasstransfer.2023.124580
5. Ronneberger, O., Fischer, P., & Brox, T. (2015). U-Net: Convolutional networks for biomedical image segmentation. In *Medical Image Computing and Computer-Assisted Intervention* (pp. 234–241). Springer. https://doi.org/10.1007/978-3-319-24574-4_28
6. ScienceDirect. (2023). A flexible deep learning framework for thermographic inspection of composites. *NDT & E Interna-tional*, 143, 102981. https://doi.org/10.1016/j.ndteint.2023.102981
7. tandfonline. (2025). Correction of emissivity in thermograms with neural networks. *Quantitative InfraRed Thermography Journal*. https://doi.org/10.1080/17686733.2025.2479949
8. arXiv. (2025). Temperature calibration of surface emissivities with an improved thermal image enhancement network (arXiv Preprint arXiv:2506.16803). https://arxiv.org/abs/2506.16803
9. Huang, H., Tang, J., Liu, T., & Huang, M. (2026). Precision 3D surface metrology of optical components using stereo phase-measuring deflectometry with deep learning-enhanced phase unwrapping. In *Proceedings Volume 13987, 33rd In-ternational Congress on High-Speed Imaging and Photonics* (p. 1398704). SPIE. https://doi.org/10.1117/12.3093993