Face mask-wearing identification systems are essential for public health monitoring, but automatic detection remains challenging for Convolutional Neural Networks (CNNs) due to facial occlusion, variations in mask types, limited datasets, and suboptimal hyperparameter selection, which limits accuracy and computational efficiency. To address this, we propose a Pelican Optimization Algorithm-based Convolutional Neural Network (POA-CNN) framework for optimized hyperparameter tuning to enhance generalization and recognition accuracy. A publicly available Kaggle dataset of 3,059 masked and unmasked face images was used. Image preprocessing was performed using Contrast Limited Adaptive Histogram Equalization (CLAHE) to improve contrast and image quality. The POA was employed to optimize critical CNN hyperparameters including number of layers, filter sizes, number of filters, and weights. The POA-CNN model was implemented in MATLAB R2023a and evaluated using accuracy, precision, sensitivity, specificity, false positive rate, and identification time, and compared against a conventional CNN. The proposed 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 identification time of 43.06 s. This outperformed the conventional CNN which achieved 94.08% accuracy, 94.67% precision, 94.91% sensitivity, 93.00% specificity, 7.00% FPR, and 65.76 s identification time. Results demonstrate that POA-CNN significantly improves classification performance while reducing computational time, making it suitable for real-time face mask-wearing identification applications.

Keywords: Pelican Optimization Algorithm, CNN, Hyperparameter Optimization, Face Mask Detection, CLAHE, Deep Learning.