Coordinate-Attention Enhanced EfficientNet for Emotion Recognition in Virtual Learning Environments

¹ Department of Computer Science Engineering (AI&ML), Vidyavardhaka College of Engineering, Mysuru, India
² Department of Electronics and Communication Engineering, Nitte Meenakshi Institute of Technology, Nitte (Deemed to be University), Bengaluru, India
³ Department of Computer Engineering, Fr. Conceicao Rodrigues College of Engineering, Mumbai, India
⁴ Department of Computer Science and Engineering, CHRIST (Deemed to be University), Bengaluru, India
⁵ Department of Humanities, Dayananda Sagar College of Engineering, Bengaluru, India

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

In recent times, emotion recognition is a continuous evolving process used to understand an individual’s emotional state especially in virtual learning environments. While existing models made improvements, a consistent challenge is to accurately recognize subtle and complex facial emotions that change rapidly, which is common with students in dynamic online sessions. To address, the proposed method CA-EffiecientNet introduces an enhanced EfficientNet architecture integrated with a Coordinate Attention (CA) block, which is specifically designed to capture both fine-grained spatial details and long-range dependencies in facial emotions. The proposed CA-EffiecientNet model improves the models ability to identify subtle signs related to a student’s emotional state such as frustration or involvement. The virtual leaning system integrates this proposed CA-EffiecientNet model to continuously monitor learners activity and dynamically adapt educational content. The performance of the CA-EffiecientNet is evaluated on the KDEF and RAF-DB datasets, achieving superior results with the accuracy of 97.6 and 98.57 compared to existing ResNet with Convolutional Block Attention Module (CBAM). The results also demonstrate the efficiency of the proposed CA-EffiecientNet model supported by its integration of MBConv layers, which reduce computational load while preserving model robustness.