Seminar Nasional Official Statistics 2020: Statistics in the New Normal: a Challenge of Big Data and Officials Statistics 370 HOW CAN MACHINE LEARNING HELP THE AUTHORITIES?: Face Mask Detection in The Era of The COVID-19 Samuel Ady Sanjaya1, Suryo Adi Rakhmawan2 1Atma Jaya University, Jakarta 2BPS-Statistics Indonesia Jakarta, Indonesia E-mail: [email protected] ABSTRACT Corona Virus Desease (COVID-19) pandemic is causing health crisis in every region in the world, especially in Indonesia. One of the effective methods against the virus is wearing face mask in public place as the regulation made by the authorities. This paper introduces face mask detection that can be used by the authorities to make mitigation, evaluation, prevention, and action planning against COVID-19. On the other hand, this solution can be used as communication tool to evaluate people’s habit on wearing face mask. The face mask recognition in this study is developed with machine learning algorithm through the image classification method: MobileNetv2. The proposed model can be integrated with surveillance camera to impede the Covid-19 transmission by allowing the detection of people who are not wearing face mask. After the training, validation, and testing phase, the model can provide the percentage of people using face mask in some cities with high accuracy. The data produced also have a strong correlation to the vigilance index of COVID-19. Keywords: Face Mask Detection, MobilenetV2, Machine Learning, COVID-19Analisis Media Sosial Twitter Tentang Pendidikan Daring Pada Masa Pandemi COVID-19 di Indonesia (Permatasari, dkk) 371 INTRODUCTION Since the declaration of the COVID-19 virus as a pandemic by WHO (Lin et al., 2020; Murray et al., 2020), efforts have been made by various parties to reduce the spread of the virus (Fadare & Okoffo, 2020). With no treatment or vaccine available, Indonesia and other countries are relying on the authorities’ interventions being implemented, for instance physical distancing and wearing face mask in the public place to impede COVID-19 transmission (Feng et al., 2020; Fund, 2020; Wilder-Smith & Freedman, 2020). Furthermore, since the New Normal has been implemented, the people are forced by law to wear face mask in the public place and wherever they interact with other people (Setiati & Azwar, 2020). There are some places in Indonesia which has regional law for using face mask in public place such as Bantul Jogjakarta, DKI Jakarta, and Provincial office of Jawa Barat set fines for residents who leave the house without wearing a mask. The second is Lebak Banten, the government punish people who do not use face mask in public place to clean public facilities which have special sign. And the last example is Banjarmasin Kalimantan Selatan, the government punish all the people who do not wear face mask in public place to do some physical punishment, such us doing push-up. However, the process of monitoring large groups of people by the government or the authorities is becoming more difficult (Loey et al., 2020). The authorities need a solution to be able to validly control the implementation of the law, which begins with the availability of the data quickly and accurately. One of the solutions is to use regionally automated face mask recognition to differentiate between people who wear masks and those who do not (Ejaz et al., 2019; Hussain & Al Balushi, 2020; Qin & Li, 2020). This paper introduces face mask detection that can be used by the NSO providing the data for the government, so the government can do some preventive action, mitigation, and evaluation of their programs. Moreover, this paper can be early warning for the authorities in capturing the people’s habit in their reginal. On the other hand, this solution can be used by the industries to provide the face mask based on the people’s habit on wearing face mask; the more people get used to wearing face mask, the more face mask need to be supplied. The proposed model can be integrated with surveillance camera to impede the Covid-19 transmission by allowing the detection of people who are not wearing face mask. Each camera point is supplied with location data, so the data can be used to determine which locations require more attention from the authorities. METHODS The face mask recognition in this study is developed with machine learning algorithm through the image classification method: MobileNetv2. Mobilenetv2 is a method based on Convolutional Neural Network (CNN) that developed by Google with improved performance and enhancement to be more efficient (Sandler et al., 2018). This study conducted its experiments on two original datasets. The first dataset was taken from the Kaggle dataset and the Real-World Masked Face dataset (RMFD); used for the training, validation, and testing phase so the model can be implemented to the dataset. The model can be produced by following some steps as shown on the Figure 1.Seminar Nasional Official Statistics 2020: Statistics in the New Normal: a Challenge of Big Data and Officials Statistics 372 Figure 1. The Step in Developing the Model The second dataset is used to apply the model to the dataset from 25 cities in Indonesia. Some cities were chosen based on data availability. The dataset was taken from some sources, for instance, public place CCTV, shop, and traffic lamp camera. Considering the quota sampling, the images were chosen based on the population proportional size of the cities, while the duration of capturing the image is equal for every city. RESULT AND DISCUSSION A. BUILDING THE FACE DETECTION MODEL 1. Data Collecting The development of the Face Mask Recognition model begins with collecting the data. The dataset train data on people who use masks and who do not. The model will differentiate between people wearing masks and not. This study uses 1.916 data with mask and 1.930 data without mask. At this step, the image is cropped until the only visible object is the face of the object. After the data has been collected, the data is labelled and grouped into two part; with mask and without mask. 2. Pre-processing The pre-processing phase is a phase before the training and testing the data. There are four
View Full Document
Unlocking...