Design and Implementation of a Deep Learning-Based “Smart Classroom Manager” System
Abstract
Verifying whether learners are actually engaged with online course resources is challenging. Ineffective learning behaviors are common. These include dropping out midway and letting videos play without watching. The "Smart Classroom Manager" is a deep learning system with three layers: perception, analysis and application. The recognition module is a convolutional neural network for facial expressions. From its outputs, a learner psychological model is derived in the two-dimensional valence–arousal space, and the results are mapped to four learning emotion labels: focus, confusion, joy, and fatigue. These labels feed three functions: facial attendance tracking, early warning of abnormal learning states, and learning behavior profiling. Online classrooms impose strict latency limits, and two engineering choices address this constraint. Facial landmark prediction shrinks the region the detector has to search, and edge computing takes part of the inference load off the server. In practice, the system lets online teaching platforms watch learning states as they happen, and the data it collects can support ongoing reforms in teaching evaluation.
