Research on Athlete Movement Recognition and Evaluation System Based on Deep Learning

This paper presents an improved gymnastics action recognition model, G-ET-GCN, which combines graph attention network (GAT) and extended temporal convolutional network (ET-CNN). G-ET-GCN effectively captures spatial and temporal information through graph attention mechanism, improving the accuracy of gymnast action recognition. Additionally, by increasing the number of layers in the extended temporal convolutional network and introducing residual mechanism, the model addresses the shortcomings of the standard TCN in temporal feature extraction, further improving its performance. Experimental results show that G-ET-GCN performs well on multiple datasets, including NTU RGB+D, SBU Kinect Interaction dataset, along with a series of gymnastics action datasets compiled by the authors, exhibits substantial enhancements in accuracy, recall, and F1 score when compared to other baseline models. Future research will explore more types of sports and more complex action recognition tasks to further validate and optimize the performance of G-ET-GCN model, driving the development of sports data analysis and action recognition technology.