Contents I. Introduction
With the booming of the emerging sports industry in China, more and more people are spending their time on sports activities such as golf and skiing [1]. For beginners, without systematic learning of sports techniques levels cannot be improved and can even lead to injuries. It is therefore essential to constantly analyse and improve the way they play sports. Traditional training methods require professional sports coaches to teach one-on-one, which is more expensive in terms of labour and less flexible. How to teach and train more simply and effectively is an issue that needs to be addressed [2]–[3]. Professional athletes often use sports assistance systems for their daily training. Athletes use motion sensors and professional training analysts to model the athlete based on sensor data, analyse data on physical oscillations during training and correct the details of the athlete [4]–[5]. Sports training systems are currently expensive to equip and require specialist motion sensors, which are neither convenient nor affordable for the average sports enthusiast. In recent years, with the availability of large amounts of data and a significant increase in parallel computing resources, deep learning has made significant progress in the fields of computer vision and natural language processing, with many practical breakthroughs in the direction of target detection, machine translation and action recognition in particular [6].
