Cellular Regression Machine: A Collaborative Modeling Approach for Batch BRDF Reconstruction and Infrared Emissivity Prediction of Non-Lambertian Surfaces

Infrared emissivity constitutes a crucial parameter for noncontact temperature measurement via infrared detectors. This article investigates noncontact measurement approaches for the infrared emissivity of non-Lambertian surfaces. A novel measurement method is put forward, which combines the merits of the integral sphere reflection method and the irradiation reflection method. By collecting the reflectance information of the target at diverse incident and exit directions, the reflection distribution in the hemispherical space is reconstructed, thereby indirectly obtaining the emissivity. The core technology is a batch regression model cellular regression machine (CRM), which employs the bidirectional reflectance distribution function (BRDF) as the kernel function. The CRM framework features a modular architecture with plug-and-play BRDF kernel compatibility, enabling reconstruction of complete BRDF distributions from the sparse reflectance data. Through rigorous Monte Carlo uncertainty analysis, we establish noise tolerance thresholds for practical deployment scenarios. An emissivity estimation model is constructed based on a simplified Cook–Torrance BRDF model, with hyperparameters optimized via comprehensive sensitivity analysis. The validity is verified through simulations demonstrating robust convergence and experiments confirming accuracy in practical applications. The computational scalability is achieved through multithreading optimization, maintaining per-cell time invariance under fixed hyperparameters. Future work includes exploring new BRDF models, implementing GPU acceleration, and extending the CRM architecture to other domains. In conclusion, this method provides new concepts and experimentally validated technologies for noncontact emissivity measurement.