Self-Supervised Temperature Representation Learning for Fever Screening.

Area of Science:

• Computer Vision
• Medical Imaging
• Machine Learning

Background:

• Thermal infrared facial imaging is crucial for public fever screening to control infectious diseases like influenza.
• A significant challenge is the scarcity of labeled fever data, hindering the development of accurate facial temperature representation learning.

Purpose of the Study:

• To propose a self-supervised fever screening framework (SelfFS) to address the data limitation in learning facial temperature representations from infrared images.

Main Methods:

• SelfFS utilizes rate reduction theory to prioritize temperature features by adjusting coding rates based on facial temperature and appearance.
• Sparsity constraints are imposed on network parameters to encourage the extraction of simple temperature features and filter out complex appearance details.

Main Results:

• The SelfFS framework demonstrates superior performance compared to existing fever screening techniques.
• It achieves results comparable to traditional supervised learning methods in fever screening.

Conclusions:

• SelfFS offers an effective self-supervised approach for fever screening using thermal infrared facial imaging, overcoming data scarcity issues.
• The method shows promise for improving public health surveillance and disease control strategies.