<![CDATA[A face mask is essential for protecting individuals from the entry of infectious or hazardous materials through the nose or mouth in specific situations. To optimize its protective function, it must be worn correctly. This research aims to develop a multiclass classification model, rather than a binary one, to assess the correctness of wearing face mask. The proposed model is designed to achieve high accuracy while maintaining efficiency, with a low number of model parameters. To this end, a deep convolutional neural network (CNN), specifically EfficientNet, is utilized. Experiments are conducted on the public MaskedFace-Net image dataset, which consists of four categories (correctly masked, uncovered chin, uncovered nose, and uncovered nose and mouth), using 3,000 randomly selected images from each category. The experiments test several EfficientNet models (B0-B3) and network hyperparameters (learning rate and dropout). The best accuracy of 0.99 is achieved by EfficientNet-B0 with a learning rate of 0.01 and a dropout rate of 0.2. The EfficientNet-B0 model outperforms other benchmark CNN models, including MobileNet-V3 and Inception-V3, despite having a slightly higher number of parameters than MobileNet-V3. This result demonstrates that the EfficientNet model is both accurate and efficient for multiclass classification of the correctness of wearing face mask.]]>
Copyrights © 2025
