Self-supervised Learning for Small Shot COVID-19 Classification

ABSTRACT

Recently, COVID-19 has become one of the most severe and widespread diseases with an increasing number of infections and deaths. An accurate and high-speed automatic classifier will increase the efficiency of diagnosis and reduce fatigue misdiagnosis. Given the contradiction that many previous classifiers require a large amount of data for training while it is difficult to collect the medical images of COVID-19 with labels, we propose a classification model based on self-supervised learning and transfer learning, which uses rotation and division as labels and then transfers the parameters to the classifier. It solves the overfitting problem caused by insufficient data set and improves the accuracy by nearly 30% © 2021 ACM.