The application of artificial intelligence and affective computing technologies in education is gaining momentum. This paper proposes a deep learning-based classroom affective guidance system that innovatively combines multimodal affective recognition with teaching strategy recommendations to achieve real-time analysis of students’ affective states and personalized teaching guidance. Through practical application in computer science courses at a vocational college, the system’s positive role in enhancing students’ learning interest and learning outcomes has been validated. The research offers new insights into building intelligent teaching environments and holds significant implications for advancing the application of affective computing in education.