In the context of college physical education curriculum reform, fostering students' interest and promoting lifelong physical exercise have become crucial. Aerobics, an integral component of physical education, plays a key role in achieving these objectives. However, existing data flow analysis technologies lack integration, limiting their ability to leverage information from various sources. To address this issue, this paper proposes an aerobics teaching model utilizing few-shot learning technology for data flow analysis. The model incorporates a label feature network based on metric learning, enhancing its ability to analyze multi-scale features and label features within classes. Comparative analysis demonstrates an 8.12% improvement in accuracy compared to traditional image feature combined classifier models.