In the context of rapid development of artificial intelligence technology, and in response to the demand for improving the quality of classroom teaching in vocational education, this paper designs a face emotion recognition model based on multiscale feature extraction. The model innovatively adopts a three-branch parallel structure to extract multi-scale features, introduces spatial and channel attention mechanisms to achieve feature adaptive fusion, and effectively improves the complex expression recognition capability. Experimental validation shows that the model has good recognition effect and real-time performance in vocational education classroom environment, which provides strong support for intelligent classroom construction and personalised teaching, and is of great significance to promote the improvement of vocational education teaching quality.