User emotion recognition and indoor space interaction design: a CNN model optimized by multimodal weighted networks

Area of Science:

• Computer Science
• Human-Computer Interaction
• Artificial Intelligence

Background:

• Intelligent user-interior interaction design requires understanding user emotions.
• Current visual emotion recognition methods using single features like facial expressions have limitations.
• Accurate visual emotion identification is crucial for effective interaction design.

Purpose of the Study:

• To develop a deep learning-based multimodal weighting network model for enhanced visual emotion recognition.
• To address the limitations of existing methods that rely on singular features.
• To improve the accuracy and coverage of visual emotion recognition for interior interaction design.

Main Methods:

• A deep learning multimodal weighting network was developed.
• The model incorporates a convolutional attention module with a self-attention mechanism within a convolutional neural network (CNN).
• A weight network classifier was derived from optimized weights for visual emotion recognition.

Main Results:

• The proposed model achieved a 77.057% correctness rate and a 74.75% accuracy rate in visual emotion recognition.
• Comparative analysis demonstrated the superiority of the multimodal weight network model over existing methods.
• The model shows potential for enhancing human-centric and intelligent indoor interaction design.

Conclusions:

• The deep learning multimodal weighting network model significantly improves visual emotion recognition.
• This advancement can lead to more sophisticated and responsive human-centric interior interaction systems.
• The model offers a promising approach for future developments in intelligent environment design.