Enhancing Facial Expression Recognition System in Online Learning Context Using Efficient Deep Learning Model

Abstract

This paper presents an online educational platform that leverages facial expression recognition technology to monitor students’ progress within the classroom. Periodically, a camera captures images of students in the classroom, processes these images, and extracts facial data through detection methods. Subsequently, students’ learning statuses are assessed using expression recognition techniques. The developed approach then dynamically refines and enhances teaching strategies using the acquired learning status data. In the course of the experiment, we enhance facial expression recognition accuracy through the utilization of ResNet-50 for effective feature extraction. Additionally, by adjusting the residual down-sampling module, we bolster the correlation among input features, thus mitigating the loss of feature information. Simultaneously, a convolutional attention mechanism module is incorporated to reduce the influence of irrelevant areas within the feature map. The proposed method achieves an accuracy of 87.62% and 88.13 % on the RAF-DB and FER2013 expression datasets, respectively. In comparison with the original ResNet-50 network and the expression recognition outcomes found in existing literature, the suggested approach demonstrates enhanced accuracy and improved detection of students’ learning states and expression variations. Consequently, the application of facial expression recognition technology in online learning, along with the optimization of online teaching resources and strategies grounded in the results of recognition, holds tangible value for augmenting the quality of online learning experiences. We have benchmarked the proposed model against state-of-the-art techniques and conducted evaluations using the FER-2013, CK+, and KDEF datasets. The significance of these results lies in their potential application within educational institutions.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 52 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.