Mask Detection Using Framework Tensorflow and Pre-Trained CNN Model Based on Raspberry Pi

Abstract

The use of masks is one of the health protocols for the community in public places and facilities in the context of preventing and controlling Covid-19. The covid-19 pandemic around the world has encouraged decision makers and various elements of society to take part in various fields and contribute to suppressing the spread of covid-19. Machine learning and computer vision are one of the branches of Artificial Intelligence (AI) and can be developed in various image recognition. In this study, a machine learning Application Program Interface (API) was used, namely Tensorflow and the pre-trained CNN model, and the Raspberry Pi, which is a mobile device as a system for detecting masks. The analysis was conducted to determine the accuracy, precision and recall (sensitivity) of the results of system implementation. The implementation results show that the use of transfer learning and fine-tuning is sufficient to help the model training process. When the model is run on the Raspberry Pi device, an accuracy percentage of 96% is produced on testing with image file input and 91% on testing with video input (realtime), 100% precision on testing with image file input and 80% on testing with video input (realtime) ). In the true positive rate (recall) performance, it was found 92% on tests with image file input and 100% on tests with video input (realtime). However, in this study, there are still type I errors in testing with video input (realtime) and type II errors in testing with image file input. In the future, it is hoped that there will be improvements and developments from this research by improving the quality of the dataset and using higher computational resources.