Political identity directly affects the implementation of the nurturing function of politics and the enhancement of the teaching effect, this paper proposes a path to enhance students’ political identity in the teaching of Daofa based on the multimodal fusion model. The TF-IDF method is utilized to identify political emotions contained in students’ classroom expressions, and LBP is used to extract students’ facial expression features. Based on the above-extracted results, a multimodal feature learner emotion data model is designed, which is also combined with a deep RNN model to identify and analyze students’ learning behaviors in the actual Daofa classroom. Evaluating students’ political identity from five dimensions, the results show that all students have significant differences in the five dimensions, P<0.05, and their political identity is above 60, of which the highest political identity in the fourth year of college reaches 78.586. Among the states of political identity, the state of maturity type has the highest frequency of occurrence, which is 132 times. The results of the study provide data support for improving students’ political identity.