Affiliations:
School of Sports and Art, Guangzhou Sports Institute, Guangzhou, Guangdong Province, China
Correspondence:
[*]
Corresponding author. Zhang Xiaolong, School of Sports and Art, Guangzhou Sports Institute, Guangzhou 510500, Guangdong Province, China. E-mail: 269878707@qq.com.
Abstract: Athletes have a large amount of video information, so how to capture effective information is the key to improving athletes’ training efficiency and improving the quality of the game. From the perspective of deep learning, this study analyzes and improves traditional algorithm models based actual needs, and jointly learns multi-scale features. At the same time, in view of the problem of over-fitting in the model training process, this study uses the sparse pyramid pool strategy to adjust the pool parameterization process and reduce the complexity of feature description. In addition, the research designs experiment to analyze the performance of the improved algorithm model and select the appropriate database to analyze the recognition effect of the algorithm model. The research shows that the algorithm of this research has a certain improvement in the recognition effect of athletes, and the recognition effect matching the artificial design features can be obtained, and it can provide theoretical reference for subsequent related research.
Keywords: Deep learning, convolution algorithm, motion recognition, database
management, deep learning
