Deep learning framework for subject-independent emotion detection using wireless signals.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Current emotion detection relies on cameras, often requiring subject-specific machine learning models.
• Existing methods lack generality and are limited in real-world applications for behavior monitoring.

Purpose of the Study:

• To develop a subject-independent method for emotion state recognition using wireless signals.
• To propose and evaluate a novel deep neural network (DNN) architecture for emotion classification from RF data.

Main Methods:

• Collected heartbeat and breathing signals from 15 participants using radio frequency (RF) reflections.
• Applied novel noise filtering techniques to the RF data.
• Developed a DNN model fusing raw and processed RF signals for emotion classification.

Main Results:

• Achieved 71.67% classification accuracy for emotion states in independent subjects.
• Obtained precision, recall, and F1-scores of 0.71, 0.72, and 0.71, respectively.
• Demonstrated superior performance of the DNN model compared to classical machine learning algorithms and validation with ECG signals.

Conclusions:

• Wireless signals, specifically RF reflections, offer a viable and accurate alternative for non-contact emotion state detection.
• The proposed DNN architecture shows significant potential for advancing behavioral science studies and well-being monitoring.