Digital intelligence for action recognition in sports dance teaching under CNN network and Bi-LSTM model

Action recognition and standardized judgment are important for sports dance training and teaching. However, existing methods remain limited in capturing fine-grained spatio-temporal features and key teaching frames, particularly for action quality assessment. Therefore, this study proposes a systematic framework for sports dance teaching evaluation based on the collaborative design of pose semantic feature extraction, temporal attention, and contrastive learning.

Specifically, biomechanically informed pose semantic features are constructed to replace the original coordinates, enabling more sensitive characterization of deviations from standard movements. Bidirectional Long Short-Term Memory (Bi-LSTM) is then used to model movement evolution, while a temporal attention mechanism automatically identifies key teaching frames. Contrastive learning further enhances the discrimination between standard and non-standard actions in the feature space, which is not directly achieved by conventional classification loss functions.

Experimental results show that the proposed model achieves an accuracy of 95.3% and an F1 score of 92.0% for action recognition, significantly outperforming other existing action recognition models. For action standard judgment, the accuracy reaches 88.5%, with reductions in both the false positive and false negative rates. Overall, the proposed method effectively integrates spatial and temporal movement information to improve the assessment of movement standardization, providing practical reference for sports dance teaching.

This study received no external funding. Physical Education College, Fujian Polytechnic Normal University, Fuqing, 350300, China Physical Education College, Xinyu University, Xinyu, 338004, China Physical Education College, Hangzhou Normal University, Hangzhou, 311121, China The authors declare no competing interests. This study was conducted in accordance with the Declaration of Helsinki.

Written informed consent was obtained from all participants for the use of their video data in this study and for the publication of anonymized images. All methods were performed in accordance with relevant guidelines and regulations. Written informed consent was obtained from all individual participants for the use of their video data in this study and for the publication of anonymized images in online open-access publications.

All images included in Figs. 9 and 11 were fully anonymized by covering the entire head region with opaque black blocks. No names or other identifying information are included in the text, figures, tables, or supplementary materials. Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.