With the increasing level of science and technology, the corresponding requirements for the current level of sports competition are also increasing. With the rapid development of international sports today, many social problems existing in sports have begun to be exposed. The rapid development of our country in this field has resulted in major achievements and many problems, such as excessive participation of athletes in self-injury training, unhealthy dietary habits and weight management, and participation in training when injured. This is a manifestation of the problem of positive deviation behavior. Compared with obvious negative deviation behavior, positive deviation behavior is more hidden, harmful and destructive. Assuming that the social issues in these sports are not well understood and studied, long-term development will not only damage the physical condition, mental state, mental stability, and sports life of athletes, but also, most importantly, it will affect the sustainable development of competitive sports. And the sound operation of society. With the rapid development of language and a deep understanding of natural things, with the deepening of research, the more traditional action recognition method is to manually extract features from the original data, which takes more time. In addition, the information conveyed by sensor signal data is much less than text and images, so it is necessary to maximize the use of pre-processing and rich knowledge data. A deep learning model can master the deep features. These features are not extracted manually by humans, but can use network models for autonomous learning, and can improve the accuracy and efficiency of recognition, and note that the algorithm can also be greatly improved Utilization of sensor data.

随着科技水平的不断提高，对当前体育竞技水平的相应要求也越来越高。在国际体育飞速发展的今天，体育中存在的许多社会问题开始暴露出来。我国在该领域的快速发展，导致了运动员过度参与自伤训练、不良饮食习惯和体重管理、受伤时参与训练等重大成就和诸多问题。这是正偏差行为问题的一种表现。与明显的负偏差行为相比，正偏差行为更具隐蔽性、危害性和破坏性。假设这些运动中的社会问题没有得到很好的理解和研究，长期的发展不仅会损害运动员的身体状况、精神状态、心理稳定性和运动生活，更重要的是会影响竞技体育的可持续发展。以及社会的良性运转。随着语言的飞速发展和对自然事物的深入理解，随着研究的深入，比较传统的动作识别方法是从原始数据中人工提取特征，耗时较多。此外，传感器信号数据传达的信息远少于文本和图像，因此需要最大限度地利用预处理和丰富的知识数据。深度学习模型可以掌握深度特征。这些特征不是人工提取的，而是可以使用网络模型进行自主学习，