Face mask-wearing identification systems leverage technology to automatically detect whether individuals are wearing face masks in public spaces. Identifying individuals wearing face masks remains a significant challenge for Convolutional Neural Networks (CNNs) due to the obstruction and occlusion of critical facial features, coupled with variations in mask types, limited datasets, and ethical concerns. However, many existing CNN techniques suffer from difficulty in selecting optimal hyperparameters, which constrain their accuracy, interpretability, and computational efficiency. Hence, this research developed a Pelican Optimization Algorithm (POA)-based Convolutional Neural Network (POA-CNN) framework to optimize CNN hyperparameters, thereby enhancing the model’s generalization capability and improving recognition accuracy. Face mask-wearing and non-mask-wearing image datasets were acquired from a publicly available Kaggle dataset comprising 3,059 images across two classes. Image preprocessing was performed using Contrast Limited Adaptive Histogram Equalization (CLAHE) to normalize contrast and enhance image quality prior to analysis. The Pelican Optimization Algorithm (POA) was employed to optimize critical CNN hyperparameters, including weights, number of layers, filter sizes, and number of filters, resulting in the developed POA-CNN model. The POA-CNN framework was used for feature extraction, training, and face mask-wearing identification and was implemented in MATLAB R2023a. The performance of the developed POA-CNN model was evaluated using accuracy, precision, sensitivity, specificity, false positive rate, and identification time and compared with the conventional CNN. The POA-CNN achieved an accuracy of 96.54%, precision of 96.95%, sensitivity of 96.95%, specificity of 96.00%, false positive rate of 4.00%, and an identification time of 43.06 seconds. In comparison, the conventional CNN achieved an accuracy of 94.08%, precision of 94.67%, sensitivity of 94.91%, specificity of 93.00%, false positive rate of 7.00%, and an identification time of 65.76 seconds. The developed POA-CNN model outperformed the conventional CNN by providing higher classification accuracy, precision, sensitivity, and specificity while reducing the false positive rate and computational time through effective optimization of CNN hyperparameters. Therefore, the POA-CNN model is suitable for real-time face mask-wearing identification applications due to its improved recognition performance and computational efficiency.

Keywords: Face Mask Detection, Convolutional Neural Network, Face Recognition, Deep Learning, Machine learning, Metaheuristic Optimization, Pelican Optimization Algorithm, Hyperparameter Optimization.