Anisotropic BRDF modeling and infrared emissivity prediction method for non-Lambertian coatings

Infrared emissivity is regarded as a key parameter in fields such as thermal radiation and aerospace remote sensing. Problems including low accuracy and missing directional information are encountered in traditional contact measurement methods (such as the integrating sphere method and the calorimetric method), while the Lambertian assumption and the isotropy assumption are often relied on in non-contact measurement, by which the directional characteristics of real materials are difficult to be handled. To address this issue, a directional emissivity inversion method named LORENet was proposed in this paper, in which a bimodal asymmetric physical model is embedded into a neural network as a prior. First, complex directional distributions were characterized by means of asymmetric broadening and multi-peak structures; then, parameter fields were generated by a parameter network under the guidance of the physical prior, and physical constraints were introduced so that non-negativity and energy conservation could be guaranteed; finally, the reflectance was integrated based on the reconstructed bidirectional reflectance distribution function (BRDF), by which high-precision emissivity inversion was completed. It is shown by the results that excellent accuracy and directional resolution capability are exhibited by this method in the modeling of non-Lambertian rough surfaces, by which strong practicability and generalization value are demonstrated.