Under the guidance of the student-centered education philosophy, understanding students’ emotions and needs is the key to improving the quality of teaching and promoting the development of personalized education. However, classroom emotion recognition faces challenges such as large numbers of people and occlusion in natural classroom environments. For this reason, this study develops student emotion recognition algorithms adapted to the classroom environment. First, the YOLOv8 model was improved by introducing the MCC module and the WIoUv3 loss function to enhance the detection accuracy and efficiency of the model. Second, a multi-channel emotion recognition network (MultiEmoNet) was developed to improve the accuracy of emotion recognition by combining facial expression, skeletal information, and environmental information. The experimental results show that the improved model performs well in various indexes and can better adapt to the classroom environment, with a view to promoting the development of classroom emotion recognition technology, providing a scientific basis for personalized teaching strategies, and enhancing the effectiveness of teaching.