HIC-DEEP: A Hierarchical Clustered Deep Learning Model for Face Mask Detection

Submission Deadline-30th July 2024
June 2024 Issue : Publication Fee: 30$ USD Submit Now
Submission Deadline-20th July 2024
Special Issue of Education: Publication Fee: 30$ USD Submit Now

International Journal of Research and Innovation in Applied Science (IJRIAS) | Volume VII, Issue V, May 2022 | ISSN 2454–6194

HIC-DEEP: A Hierarchical Clustered Deep Learning Model for Face Mask Detection

 Olugbenga S. Olukumoro1, Folurera A. Ajayi1, Adedeji A. Adebayo1, Al-Amin B. Usman1, Femi Johnson 2
1Computer Technology Department, Yaba College of Technology, Yaba, Lagos, Nigeria
2Computer Science Department, Federal University of Agriculture, Abeokuta, Ogun, Nigeria

IJRISS Call for paper

Abstract: The use of face masks is apparently not strange in these present days as conceptualized in the past due to the emergence of the Pandemic Covid-19 Corona virus. As part of the non-clinical preventive measures for the spread of this virus is the prescription and proper usage of face mask by the World health organization (WHO). In lieu of this, heads of organizations, directors of industries and individuals have adopted the “No facemask, no entry” policy in varieties of designs placed at their door posts. The state of the arts technologies has also been developed to help detect face mask non-compliant users. Whereas, the use of non-supervised machine learning approach for classifying and detecting Covid-19 facemask compliant users is not widespread. In this paper, HIC-DEEP (an un-supervised machine learning) model is proposed using a pre-trained InceptionV3 network for Kaggle database Image features vector extraction for subsequent computations of Euclidian, Spearman, and Pearson distance matrixes. The Hierarchical clustering method is then activated to identify face mask wearing faces from defiant faces. The distance algorithms all returned a perfect precision rate of 100% for the identification of faces with no face masks while an accuracy of 60%, 78% and 85% are achieved by Spearman, Pearson and Euclidian respectively for the cluster with full face mask compliance. However, the Euclidian distance algorithm returned the best overall accuracy in terms of the distance matrix with data points grouped along close proximities with unique clusters
Keywords:Face mask, InceptionV3, Deep learning, Euclidian distance, Spearman distance, Pearson distance, Covid-19
In the era of pandemics, efforts are geared towards flattening the curve as a proactive measure to either avoid community spread or taming the tide. The advent of the Covid-19 pandemic moreover has influenced several of such measures to reduce the vulnerability of people and as well reduce the stretch already placed on medical facilities across the globe [13, 15]. One of such non-pharmaceutical approaches towards safeguarding the lives of predisposed citizens is the mandatory use of face masks [18] by citizens of the world. Despite widespread compliance rate, non-compliance is likewise recorded which greatly slowed down global efforts towards flattening the curve of the virus. While non-compliance gains popularity, the entire world continues to convulse like a child under the cold hands of the ravaging pandemic. The complying public also had its fears [12] but the rate of non-compliance defies any logical reason as to why such a preventive measure could be discouraged.