Physics-Guided Deep Learning for Multiview Infrared Polarization-Based Subsurface Material Classification
IEEE Transactions on Instrumentation and Measurement · 2026 · 75 · 5014213-5014213
DOI: 10.1109/tim.2026.3711303 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Instrumentation and Measurement, 2026. Verfasst von Jianwei Lin und 5 weiteren Autor:innen. Die Arbeit nennt 35 Literaturangaben. Gefördert von: Fuzhou University.
Abstract (Originalsprache)
Infrared thermography is extensively employed in nondestructive testing (NDT). However, its defect detection performance degrades significantly when thermal diffusion contrast is weak or when the radiative properties of subsurface defects are unfavorable. To address this limitation, we propose a novel multiview polarization infrared imaging framework for high-precision discrimination of subsurface defects with diverse thermal characteristics. By exploiting the view-dependent polarization response of materials, a theoretical model is established to correlate variations in the degree of linear polarization (DoLP) with subsurface material properties under controlled heating conditions. Subsequently, a compact convolutional neural network (CNN) is designed to process fused inputs consisting of S0 and DoLP images acquired from optimally configured viewing angles. This approach enables the precise identification of embedded defects spanning materials with highly contrasting thermal conductivities to those exhibiting minimal thermal differences. Experimental validation on carbon-fiber-reinforced polymer (CFRP) specimens demonstrates classification accuracies exceeding 94 %, significantly outperforming traditional intensity-based thermography and k-means clustering. The proposed framework exhibits strong generalization capability to irregular defect morphologies while maintaining robust discrimination performance. This study provides a physics-guided and data-efficient solution for subsurface material classification in infrared NDT, and offers new insights into polarization-enhanced defect recognition.
Förderung
- Fuzhou University
Zitieren
APA
Lin, J., Yu, K., Liu, H., Huang, F., He, H., & Wu, X. (2026). Physics-Guided Deep Learning for Multiview Infrared Polarization-Based Subsurface Material Classification. IEEE Transactions on Instrumentation and Measurement, 75, 5014213-5014213. https://doi.org/10.1109/tim.2026.3711303
Vancouver
Lin J, Yu K, Liu H, Huang F, He H, Wu X. Physics-Guided Deep Learning for Multiview Infrared Polarization-Based Subsurface Material Classification. IEEE Transactions on Instrumentation and Measurement. 2026;75:5014213-5014213. doi:10.1109/tim.2026.3711303
BibTeX
@article{Lin2026,
author = {Lin, Jianwei and Yu, Kaimin and Liu, Honghui and Huang, Feng and He, Huayang and Wu, Xianyu},
title = {Physics-Guided Deep Learning for Multiview Infrared Polarization-Based Subsurface Material Classification},
journal = {IEEE Transactions on Instrumentation and Measurement},
year = {2026},
volume = {75},
pages = {5014213--5014213},
doi = {10.1109/tim.2026.3711303}
}
EUVIMED