High-Accuracy Face Mask Detection in Nuclear Environments using Ensemble Learning and XGBoost

Keeping to tight safety standards such as face masks in a very secured environment like a nuclear power plant, holds the utmost importance for both the people and the integrity of operation. This paper discusses a high-performance ensemble-based deep learning architecture that obtains face mask detection with improved precision and reliability by combining different state-of-the-art architectures. Models like Base CNN, InceptionV3, DenseNet121 and EfficientNet-B0 are used. Then individual models get trained on wide-ranging collection of face masks images across various conditions in isolation; the individual models are thereafter combined through mechanism of a weighted ensemble using their different strengths to detect preliminary a detection accuracy with around 98.91 %. The ensemble output is further optimized using the XGBoost algorithm for boosting, thus improving decision boundaries and correcting classification errors. The proposed hybrid model resulted in 99.32% accuracy, outperforming previous state-of-the-art approaches in face mask detection. The generalization ability of the proposed system in a variety of complex lighting conditions and complex settings is also significantly improved. Hence, it is extremely appropriate for nuclear facilities deployment. Our results show the power of ensemble learning and boosting in enhancing the robustness and accuracy of safety-critical computer vision systems, thus contributing to safer industrial operations.