<![CDATA[Purpose – This study aims to analyze and compare the performance of ResNet50 and MobileNetV3 for multi-class face mask classification under various environmental conditions. Design/methods/approach – ResNet50 and MobileNetV3 are trained using transfer learning for three-class face mask classification and evaluated under normal conditions and environmental variations, including illumination changes, blur, low compression, and rotation. Findings – Experimental results show that ResNet50 achieves an accuracy of 94.32% under normal conditions, slightly outperforming MobileNetV3 at 94.10%. Under environmental variations, the largest performance degradation is observed under darkening and blur conditions, while low compression and rotation have relatively minor effects. ResNet50 demonstrates higher robustness across most perturbation settings, whereas MobileNetV3 provides competitive performance with substantially better computational efficiency. Research implications/limitations – This study is limited to a controlled evaluation using synthetic environmental perturbations on a single dataset and does not consider broader dataset diversity. Therefore, the findings should be interpreted within the evaluated experimental conditions. Originality/value – This study provides a comparative analysis of model robustness under controlled environmental perturbations, highlighting the trade-off between robustness and computational efficiency for face mask classification systems.]]>
Copyrights © 2026
