Athletes' Action Recognition and Optimization Strategy Based on Deep Learning

The goal of this article is to explore the strategy of athletes' action recognition and optimization based on deep learning (DL), so as to improve the accuracy and real-time performance of action recognition and provide technical support for sports training, competition analysis and the development of intelligent sports equipment. Aiming at the challenges in the field of athlete's action recognition, a DL model based on EfficientNet is proposed. In terms of methods, a data set including track and field, gymnastics and ball games is constructed, and features are extracted and classified based on EfficientNet model. Then, by introducing the transfer learning mechanism, the pre-training model is used to speed up the training and improve the recognition accuracy. Futhermore, data enhancement technology is used to enrich the diversity of data sets and further improve the generalization ability of the model. Finally, by adjusting the model structure, the recognition effect of fine-grained actions is optimized. The experimental results show that the optimized model has achieved significant performance improvement in all kinds of sports, and the average accuracy rate is over 90%.