With the rapid development of information technology and the widespread adoption of home broadband and high-speed mobile networks, online education has become essential, especially during the COVID-19 pandemic. However, the neglect of physical education online has led to a decline in students' physical fitness. While familiarity with online learning has increased, there are higher expectations for teachers to engage with students' learning states and adjust their methods accordingly. Current online systems often focus on knowledge interaction, overlooking emotional engagement, which results in a significant "emotional lack." This paper proposes an online teaching model with emotional feedback using expression recognition technology, allowing teachers to monitor students' emotional states and adapt their teaching strategies. By implementing a convolutional neural network-based emotion recognition module, we can accurately classify students' learning emotions. The effectiveness of this module was tested through a nationwide online teaching experiment, showing high accuracy and practical value in addressing the issue of emotional disconnect in online education.