Improved heat transfer simulation and inversion approaches: Tackling low performance challenges in facial thermal modeling and thermal infrared recognition

Traditional human tissue heat-transfer models often oversimplify the heterogeneous thermal behavior of facial tissues, whereas more refined models suffer from low computational efficiency and unstable inversion. To address these limitations, we propose a discrete-element geometric model that enables non-uniform parameter allocation while preserving model simplicity. A vectorized parameterization strategy is introduced to avoid spatial discontinuities, together with physics-guided constraints derived from the heat transfer equation to ensure parameter validity. One-dimensional equations are further employed for efficient parameter initialization. Inversion experiments demonstrate lower reconstruction error, fewer abnormal parameter estimates, fewer iterations to convergence, and greater stability under noise perturbations. Simulation results show better agreement with real facial temperature distributions. Furthermore, the model is applied to thermal infrared image generation, effectively enhancing face recognition performance under cross-thermal conditions, with recognition accuracy improved by more than 10%. This work bridges heat transfer modeling and thermal infrared face recognition, enabling their effective cross-application.