Face Mask Detection using Convolutional Neural Network (CNN) to reduce the spread of Covid-19

Full text

Face Mask Detection using Convolutional Neural
Network (CNN) to reduce the spread of Covid-19
F.M. Javed Mehedi Shamrat
Department of Software Engineering
Daffodil International University
Dhaka, Bangladesh
javedmehedicom@gmail.com
Md. Masum Billah
Department of Software Engineering
Daffodil International University
Dhaka, Bangladesh
masum.swe.ndc@gmail.com
Md Saidul Islam
Department of Computer Science and Engineering
Jiangsu University of Science and Technology
Jiangsu, China
roney.orcl@gmail.com
Sovon Chakraborty
Department of Computer Science and Engineering
European University of Bangladesh
Dhaka, Bangladesh
sovonchakraborty2014@gmail.com
Md. Al Jubair
Department of Computer Science and Engineering
European University of Bangladesh
Dhaka, Bangladesh
jubair@eub.edu.bd
Rumesh Ranjan*
Department of Plant Breeding and Genetics
Punjab Agriculture University
Punjab, India
rumeshranjan@pau.edu
Abstract— The COVID-19 coronavirus pandemic is wreaking
havoc on the world's health. The healthcare sector is in a state of
disaster. Many precautionary steps have been taken to prevent the
spread of this disease, including the usage of a mask, which is
strongly recommended by the World Health Organization
(WHO). In this paper, we used three deep learning methods for
face mask detection, including Max pooling, Average pooling, and
MobileNetV2 architecture, and showed the methods detection
accuracy. A dataset containing 1845 images from various sources
and 120 co-author pictures taken with a webcam and a mobile
phone camera is used to train a deep learning architecture. The
Max pooling achieved 96.49% training accuracy and validation
accuracy is 98.67%. Besides, the Average pooling achieved 95.19%
training accuracy and validation accuracy is 96.23%.
MobileNetV2 architecture gained the highest accuracy 99.72% for
training and 99.82% for validation.
Keywords—face mask; max pooling; covid-19; average pooling;
mask detection; MobileNetV2; CNN
I. INTRODUCTION
The term "novel coronavirus" refers to a modern type of
coronavirus that has never been observed in humans before.
Coronaviruses are a form of the virus that can trigger a variety
of illnesses, from colds to life-threatening infections including
Middle East Respiratory Syndrome to Severe Acute Respiratory
Syndrome [1]. In December of this year, the first coronavirus-
infected patient was discovered. COVID-19 has been a
worldwide pandemic since that time [2]. Humans all around the
world are in precarious conditions as a consequence of the
pandemic. Every day, a huge amount of people become
contaminated with the disease and suffer as a result of it. At the
time of publication, almost 16,207,130 contaminated cases had
been reported, with 648,513 dead [3]. This statistic is gradually
growing. According to the World Health Organization (WHO),
the most frequent signs of coronavirus are fever, dry cough,
exhaustion, diarrhea, loss of taste, and smell [4]. Many
researchers and developers are working with diseases for several
years using machine learning and deep learning [5-9].
Jiang et al. [10] suggest Retina Facemask, a paradigm for
detecting the face mask that combines it with a bridge entity
elimination algorithm. The developed model includes a single-
stage detector that uses a feature pyramid network to achieve
slightly better precision and recall than the baseline result. To
address the lack of datasets, they used a learning algorithm [11],
well deep learning [12-16] methodology. Gupta et al. [17]
suggested a model implement social distance utilizing smart
communities and Intelligent Transportation Systems during the
COVID-19 pandemic (ITS). Their model called for the
installation of sensors in the city to monitor the movement of
objects in real-time, as well as the development of a data-sharing
network. Won Sonn and Lee [18] clarify how a smart city will
aid in the control of coronavirus spread in South Korea. A time-
space cartographer sped up the city's communication
monitoring, which included patient movement, transaction
background, mobile phone use, and cell phone position. CCTV
cameras in residential building hallways have been monitored in
real-time.
5th International Conference on Trends in Electronics and Informatics (ICOEI 2021)
Tirunelveli, India, 3-5, June 2021
Pre-Print

In the paper [19-22], M. Loey et al. showed the performance
of different machine learning algorithms in detecting face masks
and various purposes. In this study, three datasets are used for
feature extraction using ResNet50. For the classification
process, the decision tree algorithm, support vector machine,
and ensemble algorithms are used that gave high detection
accuracy on each dataset.
The main objective of the paper [23] is to detect a person
without a face mask and informing the authority to reduce the
spread of COVID-19. The image used in the process is captured
by CCTV cameras. After preprocessing the data, feature
extraction and classification are done using CNN. The trained
model shows an accuracy of 98.7%. The authors in the paper
[24] designed a binary face classifier to detect faces irrespective
of their alignment. In detect masks in arbitrary size input image
VGG – 16 Architecture is used for feature extraction [25]. In this
work, Gradient Descent is used for training the dataset while
Binomial Cross-Entropy is used as a loss function. M.S. Ejaz et
al. in [26], has implemented PCA for masked and non-masked
facial image detection. Viola-Jones algorithm is used in the
paper to detect face portion and at the same time, PCA to
compute Eigenface and the nearest neighbor (NN) classifier
distance is used for face recognition.
The rest of the document is formatted in the same way. This
sector consists of the most current developments in the field of
facial mask detection. The analysis technique for designing the
whole structure is outlined in Section II. Section III examines
the outcomes of the framework that has been created. Section IV
concludes with a hypothesis and shortcomings, as well as
suggestions for future work.
II. RESEARCH METHODOLOGY
CNN are a kind of deep neural network which is typically
used in deep learning to examine visual imagery. A CNN is a
Deep Learning algorithm that would take an image as input,
assign meaning to different parts of the image, and differentiate
between them. Because of their high precision, CNNs are used
for image detection [27] and identification. The CNN uses a
hierarchical model that builds a network in the shape of a funnel
and then outputs a fully-connected layer where all the neurons
are connected to each other and the data is stored. Artificial
Intelligence has made important strides in bridging the
difference between human and computer capabilities.
Researchers and enthusiasts alike operate in a number of facets
of the area to produce impressive performance. The field of
computer vision is one of several such fields. The goal of this
area is to allow machines to see and understand the environment
in the same way that humans do, and to use that information for
picture and video identification, image interpretation and
labeling, media recreation, recommendation systems, natural
language processing, and other functions are only a few
examples.
In this paper, we used three deep learning methods for face
mask detection, including Max pooling, Average pooling, and
MobileNetV2 architecture to detect the face mask. In Fig. 1 we
have displayed the entire proposed system diagram.
Fig. 1. Proposed model diagram.
A. Data Collection:
For mask detection, we used three different datasets with a
total of 1340 photographs. Using mobile cameras, webcams, and
CCTV video, another 120 photographs were taken. For detecting
masks from video used CCTV footage and Webcam, both of the
photos are in RGB. To avoid overfitting, we collected data from
different datasets and generated our datasets, the Real-World
Masked Face Dataset (RMFD) [28] and the Simulated Masked
Face Dataset (SMFD) [29], which we used for training and
testing purpose.
Fig. 2. Datasets images Samples.
B. Preprocessing and Augmentation of Data:
The images in the dataset are not all the same size, so
preprocessing was required for this study. The training of deep

learning models necessarily requires a large amount of data. We
used Keras' Image Data Generator method to resize all of the
images to 256 × 256 pixels. We normalized all images after
converting them to 256 × 256. For faster calculation, images are
converted to NumPy arrays. Increase the amount of data by
rotating, zooming, shearing, and horizontal flipping. Images are
gathered as well. The images are then resized to 128 x 128 for
passing through the second convolution layer, and then to 64 x
64 for passing through the third convolution layer.
C. Proposed Convolution Neural Network(CNN)
architecture:
For classification and image processing, CNN is used. CNN
consists of one or more convolution layers. CNN aims to find
features that are effective inside an image rather than working
with an entire image. There are several secret layers in CNN, as
well as an input layer and an output layer. In this research, we
have applied deep CNN with 3 convolution layers. Convolution
helps to get a new function by combining two mathematical
functions. Max pooling is a discretization method dependent on
samples. The aim is to reduce the complexity of an input
representation, enabling decisions to be made regarding features
found in the binned sub-regions. Our CNN model's working
process with Max pooling is depicted in Fig. 3.
Fig. 3. Three Convolution Layer with Max pooling operation.
This time, the same architecture is used for function
mapping, but with an average pooling process. The model's
activity is shown in Fig. 4. Average pooling takes the average of
all values within the picture matrix's area of interest, while Max
pooling takes the largest amount within that region. Our CNN
model initiates with Keras. Models. sequential (). In the first
hidden layer, the Relu activation feature is used, preceded by the
Max pooling process. Max pooling helps to gather significant
information and reduces the size of the images. After that, the
data is passed to the second convolution layer. Maximum
pooling is used once more to obtain the most notable
information. The obtained image matrix is then flattened and
trained. After that, the image matrix is flattened and trained.
Instead of using the Max pooling operation to observe the
model's performance, we used the Average pooling operation.
For more accurate training, Adam stochastic gradient descent
algorithms were used. We use 80% of our dataset's images for
training.
Fig. 4. Three Convolution Layer with Average pooling operation.
D. MobileNetV2 Architecture:
MobileNetV2 is a powerful image classification tool.
TensorFlow provides the image weights in MobileNetV2, a
lightweight CNN-based deep learning model. First, the
MobileNetV2 base layer is removed, and a new trainable layer
is added. The model analyzes the data and extracts the most
relevant features from our images. There are 19 bottleneck
layers in MobileNetV2 [30]. In the base model, we used
OpenCV, which is based on the ResNet-10 architecture [30]. To
detect the face and mask from an image and a video stream,
OpenCV's Caffemodel is used. The mask detecting classifier
receives the output face detected image. It allows for faster and
more accurate detection of masks in video streaming. In machine
learning, overfitting is a major problem. The Dropout layer was
used to ignore our model being overfitted with the dataset. Using
MobileNetV2 (include top=False), we were able to get rid of the
base layer. The pictures have been resized. The average pooling
operation is used with a pool size of 128 hidden layers in our
trainable model (7,7). In the secret layer, the Relu activation
function is used, and in the entire linked layer, the SoftMax
activation function is used. For better accuracy, we set a learning

rate of 0.01. The Adam stochastic gradient descent algorithm
aids in the model's comprehension of picture characteristics.
MobileNetV2 working layer depicted in Fig. 5.
Fig. 5. MobileNetV2 Architecture.
E. Evaluating performance using performance matrix:
We measured the performance of two models using
precision, recall, f1-score, and accuracy after completing the
training and testing phase. The formulas that we used are as
follows:
  
 (1)
  
 (2)
  
 (3)
    
 (4)
III. EXPERIMENT RESULT ANALYSIS
We used two datasets to detect masks from images: 1845
images from various sources and 120 co-author's photos taken
with a webcam and a mobile phone camera. The training and
validation accuracy after using the Deep CNN [31] model with
Max Pooling to reduce the dimension of our image feature map
is shown in Table I. The highest accuracy is 96.49% in training
data and 98.67% in validation data set.
TABLE I. OUTCOMES FOR DEEP CNN AFTER APPLYING MAX POOLING OF
DIFFERENT EPOCHS
Epoch
Training
Loss
Training
Accuracy
Validation
Loss
Validation
Accuracy
1
42.13%
89.76%
12.32%
90.73%
2
10.01%
91.87%
8.43%
94.34%
3
8.45%
93.97%
7.33%
96.10%
4
8.21%
94.53%
7.25%
96.25%
5
7.04%
94.98%
7.10%
97.03%
6
6.90%
95.12%
6.35%
97.23%
7
6.83%
95.24%
6.12%
97.54%
8
6.56%
95.65%
6.01%
97.71%
9
5.99%
95.89%
4.88%
97.92%
10
5.83%
96.07%
4.76%
98.12%
11
5.72%
96.45%
4.65%
98.36%
12
5.12%
96.48%
4.23%
98.43%
13
5.05%
96.49%
4.12%
98.67%
The training accuracy and validation accuracy graphs are
shown in Fig. 6. Later on, the same CNN architecture is applied
later where Average Pooling is used to reduce the dimensions of
the feature map. Compared to the previous one, the expected
outcome is less accurate. The estimated outcomes as seen in
Table II, with a maximum training accuracy of 95.19% and a
training loss of 5.92%, and a validation accuracy of 96.23%.
Fig. 6.Test Accuracy and Training Accuracy for CNN with Max Pooling
Layer
TABLE II. OUTCOMES FOR DEEP CNN AFTER APPLYING AVERAGE POOLING
OF DIFFERENT EPOCHS
Epoch
Training
Loss
Training
Accuracy
Validation
Loss
Validation
Accuracy
1
43.54%
88.92%
13.32%
89.95%
2
11.80%
90.21%
9.43%
90.12%
3
10.99%
90.85%
9.33%
91.01%
4
9.82%
91.06%
8.25%
91.52%
5
9.21%
91.24%
8.10%
93.25%
6
8.95%
92.37%
8.35%
93.54%
7
8.71%
92.69%
7.12%
94.21%
8
8.12%
94.01%
7.01%
94.39%
9
7.10%
94.29%
7.88%
95.11%
10
6.75%
94.65%
6.76%
95.15%
11
6.62%
94.82%
6.65%
95.20%
12
6.32%
95.12%
6.23%
96.12%
13
5.92%
95.19%
5.12%
96.23%
For each epoch, Fig. 7 depicts a graph of relative validation
and training accuracy.

Fig. 7.Test Accuracy and Training Accuracy for CNN with Average
Pooling Layer.
The accuracy improved significantly by using the
MobileNetV2 architecture. For each epoch, Table III shows the
validation and test accuracy.
TABLE III. DIFFERENT OUTCOMES AFTER APPLYING MOBILENETV2
ARCHITECTURE.
Epoch
Training
Loss
Training
Accuracy
Validation
Loss
Validation
Accuracy
1
4.43%
98.67%
4.21%
98.71%
2
4.32%
98.72%
4.12%
98.81%
3
4.21%
98.81%
4.09%
98.92%
4
4.12%
98.92%
3.89%
99.10%
5
3.90%
98.99%
3.72%
99.13%
6
3.82%
99.01%
3.61%
99.25%
7
3.78%
99.13%
3.56%
99.32%
8
3.65%
99.24%
3.41%
99.37%
9
3.61%
99.32%
3.23%
99.47%
10
3.54%
99.51%
3.20%
99.65%
11
3.46%
99.63%
3.18%
99.82%
12
3.42%
99.72%
3.18%
99.82%
13
3.42%
99.72%
3.18%
99.82%
The best precision is 99.72% for training data and 99.82
percent for validity data, according to Table III. Just 3.18% of
data lost during the validation process. Fig. 5 shows the detailed
comparison of test accuracy and validation accuracy of
MobilenetV2 which is a CNN-based architecture. After using
the MobilenetV2 architecture, we measured the confusion
matrix. The confusion matrix is correctly depicted in Fig. 8.
Fig. 8.Confusion Matrix after applying MobilenetV2.
The MobilenetV2 design outperformed many of the other
models included in this study. This model is capable of
recognizing the mask in a picture. In Fig.9, 10, and 11 showing
the detection result of MobileNetV2.
Fig. 9.Detection of No Mask from an image.
Fig. 10.Detection of Mask from an image.

MobilenetV2 can successfully identify the mask from video
streams with proper accuracy.
Fig. 11. Detection of the mask from video streams using MobilenetV2.
The Max pooling achieved 96.49% training accuracy and
validation accuracy is 98.67%. Besides, the Average pooling
achieved 95.19% training accuracy and validation accuracy is
96.23%. MobileNetV2 architecture gained the highest accuracy
99.72% for training and 99.82% for validation. A short
explanation is added in Table IV.
TABLE IV. COMPARISON WITHIN THE CNN TECHNIQUES
Epoch
Training
Loss
Training
Accuracy
Validation
Loss
Validation
Accuracy
Max Pooling
13
5.05%
96.49%
4.12%
98.67%
Average
Pooling
13
5.92%
95.19%
5.12%
96.23%
MobileNetV2
13
3.42%
99.72%
3.18%
99.82%
IV. CONCLUSION AND FUTURE WORK
We used two deep CNN architectures and one CNN-based
MobilenetV2 architecture in this study. Our primary objective
was to propose a compatible model with high accuracy such that
mask identification will be simple throughout the pandemic. In
order to assess performance with a wider dataset, we can attempt
to add further models to compare with Mobilenetv2 and tried to
integrate this model with IoT [32-35] to detect humans without
masks automatically.
REFERENCES
[1] WHO EMRO | About COVID-19 | COVID-19 | Health topics.
[Online]. Available: http://www.emro.who.int/health-
topics/corona-virus/about-covid-19.html, accessed on: Jul. 26,
2020.
[2] H. Lau et al., “Internationally lost COVID-19 cases,” J. Microbiol.
Immunol. Infect., vol. 53, no. 3, pp. 454–458, 2020.
[3] Worldometer, “Coronavirus Cases,”. [Online]. Available:
https://www.worldometers.info/coronavirus, accessed on: Jul. 26,
2020.
[4] L. Li et al., “COVID-19 patients’ clinical characteristics, discharge
rate, and fatality rate of meta-analysis,” J. Med. Virol., vol. 92, no.
6, pp. 577–583, Jun. 2020.
[5] P. Ghosh et al., "Efficient Prediction of Cardiovascular Disease
Using Machine Learning Algorithms With Relief and LASSO
Feature Selection Techniques," in IEEE Access, vol. 9, pp. 19304-
19326, 2021, doi: 10.1109/ACCESS.2021.3053759.
[6] F.M. Javed Mehedi Shamrat, Md. Asaduzzaman, A.K.M. Sazzadur
Rahman, Raja Tariqul Hasan Tusher, Zarrin Tasnim “A
Comparative Analysis of Parkinson Disease Prediction Using
Machine Learning Approaches” International Journal of Scientific
& Technology Research, Volume 8, Issue 11, November 2019,
ISSN: 2277-8616, pp: 2576-2580.
[7] A.K.M Sazzadur Rahman, F. M. Javed Mehedi Shamrat, Zarrin
Tasnim, Joy Roy, Syed Akhter Hossain “A Comparative Study on
Liver Disease Prediction Using Supervised Machine Learning
Algorithms” International Journal of Scientific & Technology
Research, Volume 8, Issue 11, November 2019, ISSN: 2277-8616,
pp: 419-422.
[8] F. M. Javed Mehedi Shamrat, Md. Abu Raihan, A.K.M. Sazzadur
Rahman, Imran Mahmud, Rozina Akter, “An Analysis on Breast
Disease Prediction Using Machine Learning Approaches”
International Journal of Scientific & Technology Research, Volume
9, Issue 02, February 2020, ISSN: 2277-8616, pp: 2450-2455.
[9] F. M. Javed Mehedi Shamrat, Zarrin Tasnim, Imran Mahmud, Ms.
Nusrat Jahan, Naimul Islam Nobel, “Application Of K-Means
Clustering Algorithm To Determine The Density Of Demand Of
Different Kinds Of Jobs”, International Journal of Scientific &
Technology Research, Volume 9, Issue 02, February 2020, ISSN:
2277-8616, pp: 2550-2557.
[10] M. Jiang, X. Fan, and H. Yan, “RetinaMask: A Face Mask detector,”
2020. [Online]. Available: http://arxiv.org/abs/2005.03950.
[11] P. Ghosh, S. Azam, A. Karim, M. Jonkman, MDZ Hasan, “Use of
Efficient Machine Learning Techniques in the Identification of
Patients with Heart Diseases,” 5th ACM International Conference
on Information System and Data Mining (ICISDM2021), 2021.
[12] M. S. Junayed, A. A. Jeny, S. T. Atik, N. Neehal, A. Karim, S.
Azam, and B. Shanmugam, “AcneNet - A Deep CNN Based
Classification Approach for Acne Classes,” 2019 12th International
Conference on Information & Communication Technology and
System (ICTS), 2019.
[13] A. Karim, P. Ghosh, A. A. Anjum, M. S. Junayed, Z. H. Md, K. M.
Hasib, and A. N. Bin Emran, “A Comparative Study of Different
Deep Learning Model for Recognition of Handwriting Digits,”
SSRN Electronic Journal, 2021.
[14] M. Al Karim, A. Karim, S. Azam, E. Ahmed, F. De Boer, A. Islam,
and F. N. Nur, “Cognitive Learning Environment and Classroom
Analytics (CLECA): A Method Based on Dynamic Data Mining
Techniques,” Innovative Data Communication Technologies and
Application, pp. 787–797, 2021.
[15] Chen, Joy Iong Zong, and S. Smys. Social Multimedia Security and
Suspicious Activity Detection in SDN using Hybrid Deep Learning
Technique. Journal of Information Technology 2, no. 02 (2020):
108-115.
[16] Smys, S., Joy Iong Zong Chen, and Subarna Shakya. Survey on
Neural Network Architectures with Deep Learning. Journal of Soft
Computing Paradigm (JSCP) 2, no. 03 (2020): 186-194.
[17] M. Gupta, M. Abdelsalam, and S. Mittal, “Enabling and Enforcing
Social Distancing Measures using Smart City and ITS
Infrastructures: A COVID-19 Use Case,” 2020. [Online]. Available:
https://arxiv.org/abs/2004.09246.
[18] J. Won Sonn and J. K. Lee, “The smart city as time-space
cartographer in COVID-19 control: the South Korean strategy and
democratic control of surveillance technology,” Eurasian Geogr.
Econ., pp. 1–11, May. 2020.
[19] Loey M, Manogaran G, Taha MHN, Khalifa NEM. A hybrid deep
transfer learning model with machine learning methods for face
mask detection in the era of the COVID-19 pandemic. Measurement
: Journal of the International Measurement Confederation. 2021
Jan;167:108288. DOI: 10.1016/j.measurement.2020.108288.
[20] F. M. Javed Mehedi Shamrat, P. Ghosh, M. H. Sadek, M. A. Kazi
and S. Shultana, "Implementation of Machine Learning Algorithms
to Detect the Prognosis Rate of Kidney Disease," 2020 IEEE
International Conference for Innovation in Technology (INOCON),

Bangluru, India, 2020, pp. 1-7, doi:
10.1109/INOCON50539.2020.9298026.
[21] P. Ghosh, F. M. Javed Mehedi Shamrat, S. Shultana, S. Afrin, A. A.
Anjum and A. A. Khan, "Optimization of Prediction Method of
Chronic Kidney Disease Using Machine Learning Algorithm," 2020
15th International Joint Symposium on Artificial Intelligence and
Natural Language Processing (iSAI-NLP), Bangkok, Thailand,
2020, pp. 1-6, doi: 10.1109/iSAI-NLP51646.2020.9376787.
[22] F. M. Javed Mehedi Shamrat, Z. Tasnim, P. Ghosh, A. Majumder
and M. Z. Hasan, "Personalization of Job Circular Announcement
to Applicants Using Decision Tree Classification Algorithm," 2020
IEEE International Conference for Innovation in Technology
(INOCON), Bangluru, India, 2020, pp. 1-5, doi:
10.1109/INOCON50539.2020.9298253.
[23] M. M. Rahman, M. M. H. Manik, M. M. Islam, S. Mahmud and J. -
H. Kim, "An Automated System to Limit COVID-19 Using Facial
Mask Detection in Smart City Network," 2020 IEEE International
IOT, Electronics and Mechatronics Conference (IEMTRONICS),
Vancouver, BC, Canada, 2020, pp. 1-5, doi:
10.1109/IEMTRONICS51293.2020.9216386.
[24] T. Meenpal, A. Balakrishnan and A. Verma, "Facial Mask Detection
using Semantic Segmentation," 2019 4th International Conference
on Computing, Communications and Security (ICCCS), Rome,
Italy, 2019, pp. 1-5, doi: 10.1109/CCCS.2019.8888092.
[25] F. M. Javed Mehedi Shamrat, Imran Mahmud, A.K.M Sazzadur
Rahman, Anup Majumder, Zarrin Tasnim, Naimul Islam Nobel,“A
Smart Automated System Model For Vehicles Detection To
Maintain Traffic By Image Processing” International Journal of
Scientific & Technology Research, Volume 9, Issue 02, February
2020, ISSN: 2277-8616, pp: 2921-2928.
[26] A. Islam Chowdhury, M. Munem Shahriar, A. Islam, E. Ahmed, A.
Karim, and M. Rezwanul Islam, “An Automated System in ATM
Booth Using Face Encoding and Emotion Recognition Process,”
2020 2nd International Conference on Image Processing and
Machine Vision, 2020.
[27] M.S. Ejaz, M.R. Islam, M. Sifatullah, A. SarkerImplementation of
principal component analysis on masked and non-masked face
recognition 2019 1st International Conference on Advances in
Science, Engineering and Robotics Technology (ICASERT) (2019),
pp. 15, 10.1109/ICASERT.2019.8934543
[28] https://github.com/X-zhangyang/Real-World-Masked-Face-
Dataset.
[29] https://www.kaggle.com/omkargurav/face-mask-dataset
[30] An automated System to limit covid 19 using facial mask detection
in smart city network( 2020, IEEE)
https://ieeexplore.ieee.org/document/9216386.
[31] Junayed M.S., Jeny A.A., Neehal N., Atik S.T., Hossain S.A. (2019)
A Comparative Study of Different CNN Models in City Detection
Using Landmark Images. In: Santosh K., Hegadi R. (eds) Recent
Trends in Image Processing and Pattern Recognition. RTIP2R 2018.
Communications in Computer and Information Science, vol 1035.
Springer, Singapore. https://doi.org/10.1007/978-981-13-9181-
1_48
[32] Javed Mehedi Shamrat F.M., Allayear S.M., Alam M.F., Jabiullah
M.I., Ahmed R. (2019) A Smart Embedded System Model for the
AC Automation with Temperature Prediction. In: Singh M., Gupta
P., Tyagi V., Flusser J., Ören T., Kashyap R. (eds) Advances in
Computing and Data Sciences. ICACDS 2019. Communications in
Computer and Information Science, vol 1046. Springer, Singapore.
https://doi.org/10.1007/978-981-13-9942-8_33.
[33] Shamrat F.M.J.M., Nobel N.I., Tasnim Z., Ahmed R. (2020)
Implementation of a Smart Embedded System for Passenger Vessel
Safety. In: Saha A., Kar N., Deb S. (eds) Advances in
Computational Intelligence, Security and Internet of Things.
ICCISIoT 2019. Communications in Computer and Information
Science, vol 1192. Springer, Singapore.
https://doi.org/10.1007/978-981-15-3666-3_29.
[34] F. M. Javed Mehedi Shamrat, Zarrin Tasnim, Naimul Islam Nobel,
and Md. Razu Ahmed. 2019. An Automated Embedded Detection
and Alarm System for Preventing Accidents of Passengers Vessel
due to Overweight. In Proceedings of the 4th International
Conference on Big Data and Internet of Things (BDIoT'19).
Association for Computing Machinery, New York, NY, USA,
Article 35, 1–5. DOI:https://doi.org/10.1145/3372938.3372973.
[35] F.M. Javed Mehedi Shamrat, Shaikh Muhammad Allayear and Md.
Ismail Jabiullah "Implementation of a Smart AC Automation
System with Room Temperature Prediction", Journal of the
Bangladesh Electronic Society, Volume 18, Issue 1-2, June-
December 2018, ISSN: 1816-1510, pp: 23-32.