Swimming-assisted training and physical fitness enhancement system based on improved YOLOv5 and improved ST-GCN
Yang 阳 Chen 陈 · International Journal of Information and Communication Technology · 2026
This paper proposes a swimmer action recognition model combining an improved YOLOv5 algorithm and an improved ST-GCN, and develops a swimming-assisted training system based on this model.The system comprises four key modules: posture recognition, data visualisation, training feedback, and physical fitness assessment.Experimental results demonstrate that the proposed model achieves 98.78% precision, 98.13% average accuracy, and 97.58% recall, outperforming comparison models.The system exhibits superior performance with a mean recognition time of 79.6 ms and CPU occupation of 42.57%.Practical application evaluation shows significant advantages in improving training effectiveness, with 237 requests per second throughput, 12.41% memory usage, and coach/athlete satisfaction rates of 98.76% and 98.47%, respectively.