[3b1165-195] コーチング学（奇数演題）／ポスター発表

[09コ-ポ-43] Do Go players have distinct playing styles? Analyzing player styles using an automatic image recognition systemGo players' playing styles

*Lung Hung Chen¹, Tsai-Cheng Wu² (1. National Taiwan Sport University, Taiwan, 2. National Tsing Hua University, Taiwan)

  Purpose: This study aims to develop a system capable of automatically recognizing and analyzing online Go matches to evaluate players' playing styles. Methods: The research team selected publicly available match videos from YouTube as the data source and developed an image recognition module. First, web scraping techniques and the YouTube API were employed to obtain the videos, and OpenCV was used to convert these videos into individual image frames. Subsequently, the board information was extracted, and Go stones were recognized using DETR and ViT models. The board information across different time sequences was then linked and integrated to achieve a coherent record of each match. The extracted data was classified and processed by the research team, who developed a metric to categorize playing styles into three distinct types. Results: The research team randomly selected approximately six images from each video for manual review to verify accuracy. A stringent accuracy formula was applied to evaluate the correctness of the image recognition module. Based on the validation results, the recognition module achieved an accuracy rate of 82.56%. Furthermore, the calculated metric was applied to the data of players Xu Haohong, Wang Yuanjun, and Lin Junyan to determine their respective playing styles. The results indicated that players Xu Haohong and Lin Junyan belong to the steady style, while Wang Yuanjun belongs to the defensive style. Conclusion: Although there is room for improvement in the overall functionality of the system, The research team successfully developed an automated method for extracting and recognizing Go board information from match videos and proposed a set of metrics to infer players' styles based on the extracted data.