Hybrid deep convolutional model-based emotion recognition using multiple physiological signals.

Area of Science:

• Affective computing
• Human-computer interaction
• Signal processing

Background:

• Emotion recognition is vital across medical, advertising, and military fields.
• Extracting accurate emotional cues from complex sensor data presents challenges.
• Advanced feature engineering is crucial for reliable signal recognition.

Purpose of the Study:

• To develop a hybrid affective model for emotion classification using wearable sensor data.
• To leverage transfer learning and signal fusion for enhanced affect recognition.
• To compare the proposed model's performance against traditional and deep learning methods.

Main Methods:

• A hybrid deep model incorporating Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Recurrent Neural Network (RNN) was developed.
• Transfer learning was applied to a genuine dataset of fused signals from 30 participants using wearable sensors and mobile devices.
• The model processed large-frame sensor signals, integrating traditional feature extraction with deep learning algorithms.

Main Results:

• The hybrid deep model achieved an average classification accuracy of 93%.
• This significantly outperformed traditional neural networks (54% accuracy) and existing deep learning approaches (76% accuracy).
• The model also surpassed a previously proposed autoregressive hidden Markov model (AR-HMM) approach (88.6% accuracy).

Conclusions:

• Deep learning methods are effective for affect recognition, especially with large frame inputs.
• The proposed hybrid deep model demonstrates superior performance in emotion classification from sensor signals.
• This advanced model offers a promising solution for accurate and reliable emotion recognition in real-world applications.