Deteksi Masker Menggunakan Model Single-Shot Multibox Detector (SSD) MobileNetV2 Dan Feature Pyramid Network (FPN

Abstract

The use of masks is one of the mitigations recommended by WHO in controlling the spread of COVID-19. However, monitoring the use of masks is less effective if only done directly. Therefore, the use of deep learning technology can be used as an effort to detect the use of masks automatically without requiring direct supervision. This study aims to develop a CNN-based model to detect the use of masks using the Single-Shot Multibox Detector (SSD) architecture. The type of model to be used is the SSD MobileNetV2 model and combined with the Feature Pyramid Network (SSD MobileNetV2 FPNLite). The SSD MobileNetV2 FPNLite model will also be compared with SSD MobileNetV2 model without FPN to determine which model is better based on accuracy and detection speed. The dataset used is a combination of the MOXA3K and MAFA datasets which consists of 15,000 images of people wearing masks and also those who are not wearing masks. To produce good accuracy, the model is also trained using a combination of hyperparameters and also using data augmentation techniques. The results showed that SSD MobileNetV2 FPNLite model could produce higher accuracy with a value of mAP, mAP@50 and mAP@75 each of 46.23%; 76.95%; and 51.59% with an inference time of 30.85 seconds to process 500 images with an average inference time of 0.061 seconds for each image.