A Method for Recognizing Students' Concentration Based on Self-attention Mechanism

Existing student concentration recognition method has two kinds of manual and automatic, manual recognition is time-consuming and laborious and easy to be affected by uncontrollable factors, automatic recognition has the advantages of being widely used and not affecting the learning in the video-based method, but the existing video recognition model has many parameters, poor ability to capture the features from a distance, and can not be parallel computation, and most of them are a combination of local and overall features, and the dataset also has the problems such as small data data volume and easy to be obscured in the classroom scene. Aiming at the need to balance the teaching process and outcome evaluation in teaching evaluation reform [9], this study proposes an automated solution based on students' classroom attention. By simulating occlusion data enhancement and random frame sampling to increase data diversity, improving ResNet to achieve multi-scale feature extraction, optimizing the self-attention mechanism to incorporate time-dimensional coding, and constructing an efficient recognition model with parallel computing [3].