To explore the application of dynamic analysis algorithm in sports training injury risk assessment, this paper takes the Spatio-Temporal Graph Convolutional Network (ST-GCN) as the main algorithm, and introduces the Adaptive Graph Convolution Module (AGCM) and Residual Channel Attention Module (RCAM). ST-GCN is improved to form AGCM + RCAM-ST-GCN (ARST-GCN) motion posture recognition algorithm. Meanwhile, combined with the extreme gradient boosting (XG Boost), the final physical training injury risk assessment model is formed. The performance of the improved ARST-GCN and the proposed damage risk assessment model is verified by experiments. The results show that ARST-GCN, which combines AGCM and RCAM modules, performs best in all indicators. Compared with ST-GCN, the accuracy rate is increased by 1.94% and the F1 value is increased by 4.3%. In addition, in the performance comparison of different sports injury risk models, the recall rate and F2 value of XGBoost are 0.937 and 0.893, respectively, and the overall performance is the best, indicating that XGBoost has significant advantages in dealing with sports injury risk assessment (SIRA) tasks. The research results provide theoretical basis and practical reference for injury prevention in sports training, and help to improve the accuracy and reliability of SIRA.
Read full abstract