A Deep Learning Approach for Fear Recognition on the Edge Based on Two-Dimensional Feature Maps.

Area of Science:

• Affective computing
• Biomedical engineering
• Machine learning

Background:

• Real-time detection of fear is crucial for personal safety.
• Wearable devices offer a non-invasive method for collecting physiological data.
• Integrating affective computing with signal processing can enhance user security.

Purpose of the Study:

• To develop and validate a fear recognition method using physiological signals from wearable devices.
• To explore the application of deep learning models inspired by image processing for affective computing.
• To assess the feasibility of edge deployment for real-time fear detection.

Main Methods:

• Physiological signals were collected from wearable devices.
• Two-dimensional feature maps were created from raw signals.
• Deep learning classification models were employed, adapted from image processing techniques.
• Data augmentation and feature selection algorithms were utilized.

Main Results:

• The proposed method achieved high performance across multiple datasets (WEMAC, WESAD 3-class, WESAD 2-class).
• F1-scores ranged from 78.13% to 99.60%, and accuracy ranged from 79.90% to 99.60%.
• Successful implementation on a Coral Edge TPU device demonstrated edge inference capabilities.

Conclusions:

• The developed fear recognition system is effective and accurate.
• The method shows promise for real-time bodyguard applications using wearable technology.
• Edge computing enables practical, on-device deployment for immediate threat detection.