The vehicle cabin represents a unique thermal environment where excessively high temperatures strain the driver's thermoregulatory system, significantly impairing concentration and reaction times. With the growing automation of modern vehicles and their reliance on intelligent systems, maintaining optimal cabin temperature has become increasingly important. In this study, a deep learning–based facial image recognition model is introduced to evaluate thermal fatigue and comfort of road vehicle driver. Specifically, a ResNet50 architecture integrated with a spatial attention mechanism (Attention-ResNet50) is designed to enhance feature extraction from facial imagery by prioritizing physiologically informative regions. To collect data, 20 participants completed driving trials on a climate-controlled driving simulator. The model training data includes facial images, thermal perception data collection form, and environmental measurements under summer and winter conditions. Each facial sample was annotated based on fatigue status (alertness or fatigue), gender and seasonal conditions. Finally, the model prediction results were compared with the thermal perception data collection form to evaluate fatigue and comfort. The verification accuracy of both models exceeds 95%. While larger training datasets improved validation accuracy by approximately 5%, gains eventually plateaued. Notably, the spatial attention mechanism significantly bolstered model performance, particularly for male drivers and summer scenarios: accuracy improved by approximately 1%, while the number of training epochs required for convergence was reduced by an average of 3 rounds. These results highlight the potential of attention-enhanced deep learning models for driver monitoring applications. The approach offers valuable prospects for adaptive climate control and real-time safety alerts, contributing to improved passenger comfort and overall driving safety. • Deep learning–based facial image recognition model detects driver thermal state • Data from 20 subjects under varied thermal and seasonal conditions • Attention mechanism improves the performance of the model • Accuracy increases as the data size grows. • Male and summer images benefit most from attention-based modeling
Li et al. (Sun,) studied this question.