Performance Measurement of ResNet-34 in Convolutional Neural Network Method for Classification of Mask Type Usage

Outbreaks of new viruses have been documented in history, as evidenced by SARS (Severe Acute Respiratory Syndrome) in 2002–2003 and MERS (Middle East Respiratory Syndrome) in 2012. This virus also belongs to the coronavirus family, which includes the COVID-19 virus. Even though Covid-19 cases have started to decrease, masks are still recommended. The application can be implemented in hospitals or places with a high probability of spreading the virus. This study used the Convolutional Neural Network (CNN) approach to classify the types of masks. Several architectures on CNN that can be used include MobileNetV2, EfficientNetB4, ResNet-18, and ResNet-34. Based on the performance comparison results obtained using the face-mask dataset taken from the Kaggle site and data taken in real time, totalling 8,641, ResN et-34 is the model with the best accuracy. The learning rate is 0.001, and the classifier function used is softmax with an input size of 229×229, cross-entropy loss function, epoch 15, and the ReLu activation function. The developed model achieves an accuracy rate of 0.9955 during training and 0.8748 during validation. So that the CNN method in the implementation of Deep Learning can classify images correctly using the type of mask.