Hybrid Model Transfer Learning ResNet50 and Support Vector Machine for Face Mask Detection

Abstract

The Covid-19 virus caused a health crisis in Indonesia. This virus is so deadly that it has caused many fatalities which have caused the whole world including the government to pay major attention to the Covid-19 pandemic. The Indonesian government has issued several policies to prevent the spread of this epidemic, one of which is wearing a mask in public places. One approach that is widely used in the field of computer vision is the Convolutional Neural Network (CNN) transfer learning. In this study, Hybrid Model Transfer Learning ResNet50 and SVM with RGB to HSV preprocessing is presented to detect masks in facial images. This model consists of three process components. The first is preprocessing RGB images to HSV, the second component is for Feature Extraction with ResNet50 and the third is mask classification on face images with Support Vector Machine (SVM). From dataset of 7328 training and testing data were carried out. The first model, without preprocessing the image data with ResNet50, produces an accuracy of 86.52%. The second model, the model with preprocessing converts image data from RGB to HSV with ResNet50 resulting in an accuracy of 99.18%. In the third model, without preprocessing with ResNet50 and SVM which has an accuracy of 90.55%. The fourth model, the model with preprocessing converts image data from RGB to HSV with ResNet50 and SVM resulting in an accuracy of 98.36%.