Spatial-Temporal Recurrent Neural Network for Emotion Recognition.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Signal Processing

Background:

• Emotion recognition is crucial for human-computer interaction.
• Existing methods often struggle to effectively integrate spatial and temporal signal features.
• Deep learning offers potential for improved feature extraction in complex signal data.

Purpose of the Study:

• To propose a novel deep learning framework, the spatial-temporal recurrent neural network (STRNN).
• To integrate spatial and temporal information from signal sources into a unified model.
• To enhance emotion recognition accuracy by leveraging spatial-temporal dependencies.

Main Methods:

• Developed a two-layer recurrent neural network (RNN) architecture (STRNN).
• Employed a multidirectional RNN layer to capture spatial co-occurrences within temporal slices.
• Utilized a bi-directional temporal RNN layer to learn discriminative temporal features.
• Implemented sparse projection to enhance model discriminant ability by selecting salient regions.

Main Results:

• The STRNN framework effectively integrates spatial and temporal signal information.
• The model demonstrated superior performance in emotion recognition tasks.
• Experimental results on electroencephalogram and facial expression datasets confirmed competitive results over state-of-the-art methods.

Conclusions:

• The proposed STRNN model offers an effective approach for emotion recognition.
• Integrating spatial and temporal dependencies significantly improves model performance.
• STRNN provides a competitive and robust method for analyzing complex signal data for emotion detection.