An Improved Lightweight Method for Micro-Expression Recognition in Online Learning

With the increasing prevalence of online learning, monitoring and evaluating students' learning status has become a central concern in the field of education. This paper proposes a network fusion-based micro-expression recognition model, aimed at analyzing students' micro-expressions to identify their online learning status and provide a basis for teachers' evaluations. The model is based on the lightweight MobileViT network and integrates the Depthwise Separable Convolution module and the Coordinate Attention mechanism, forming the integrated MobileViT-DPCA network. Additionally, we selected a video of students learning online. We captured one image every 2 seconds and filtered the image categories through self-reporting. This resulted in a Self-constructed dataset, which we used for training the model in online learning. Experimental results show that the improved network outperforms the baseline MobileViT on the CASME II dataset, SAMM dataset and Self-constructed dataset, showing accuracy improvements of 11.11%, 2.49%, and 3.88%, respectively. Finally, we validated the applicability of the model in online learning.