A Student Facial Expression Recognition Model Based on Multi-Scale and Deep Fine-Grained Feature Attention...

Area of Science:

• Artificial Intelligence
• Computer Vision
• Educational Technology

Background:

• Accurate student facial expression recognition is vital for adaptive teaching in smart classrooms.
• Existing methods struggle with precise facial feature extraction and robustness in classroom settings.

Purpose of the Study:

• To develop an advanced student facial expression recognition model (SFER-MDFAE).
• To improve the accuracy and robustness of facial expression analysis in smart classroom environments.

Main Methods:

• Proposed a multi-scale dual-pooling feature aggregation module for comprehensive facial information capture.
• Designed a key region-oriented attention mechanism to enhance fine-grained facial expression features.
• Fused multi-scale and attention-enhanced features for improved facial key information representation.

Main Results:

• Achieved high accuracy on benchmark datasets: 76.18% (FER2013), 92.75% (FERPlus), 92.93% (RAF-DB), 67.86% (AffectNet).
• Demonstrated superior performance on a real smart classroom dataset (SCFED) with 93.74% accuracy.
• Outperformed existing state-of-the-art methods in facial expression recognition tasks.

Conclusions:

• The SFER-MDFAE model effectively addresses limitations in facial feature extraction and recognition robustness.
• The proposed method significantly enhances the accuracy of student facial expression recognition in smart classrooms.
• Validated effectiveness through comprehensive experiments on diverse datasets.