通过上述阶段处理过后的数据是运动员所作动作的加速度和角速度可是这不能准确的代表所作动作的特征在进行分类时需要的是动作的特征数据所以必须从采集的原始数据中经过特征提取得到特征数据。特征数据一般分为时域特征和频域特征：1时域特征：时域是描述数学函数或物理信号对时间的变化关系。例如一个信号随着时间变化的情况可以从信号的时域波形中得到。当从一段连续的窗口数据中提取一些在数学中常见的统计特征那么这些特征称为

Abstract：In order to accurately recognize the actions of athletes, it is necessary to extract feature data from the original data collected. Feature data is generally divided into time-domain features and frequency-domain features. Time-domain features refer to the statistical features extracted from a segment of continuous window data, while frequency-domain features represent the characteristics of signal data in the frequency domain. The two types of features need to be analyzed simultaneously in order to obtain a comprehensive understanding of the signal. However, the more features extracted, the higher the complexity of the classifier and the longer the computation time required, which affects the real-time performance of the recognition system. Therefore, it is necessary to select some features according to certain rules to reduce the redundancy between different types of features, so as to improve the accuracy of recognition and the speed of the system.

Keywords: action recognition; feature extraction; time-domain features; frequency-domain features.

Introduction：With the development of wearable technologies, motion sensors have been widely used in action recognition, which is of great significance for sports training and medical rehabilitation. However, the raw data collected from motion sensors cannot accurately represent the features of the actions performed by athletes. Therefore, it is necessary to extract feature data from the original data to achieve accurate recognition. This paper mainly introduces the extraction of time-domain features and frequency-domain features, and discusses the importance of feature selection in improving the accuracy and speed of recognition.

Time-domain features：Time-domain is a mathematical representation of the relationship between a function or physical signal and time. The statistical features extracted from a segment of continuous window data are called time-domain features. These features can be directly extracted from the window data, which is relatively simple and convenient. Time-domain features mainly include mean, standard deviation, root mean square, maximum value, minimum value, and so on. These features can reflect the overall trend and variability of the motion data, and are often used as the basic features for action recognition.

Frequency-domain features：Frequency-domain refers to the part of a function or signal that is related to frequency rather than time. Frequency-domain features can represent the characteristics of signal data in the frequency domain. The Fourier transform can be used to convert the time-domain signal into a frequency-domain signal. Frequency-domain features mainly include power spectral density, spectral centroid, spectral flatness, and so on. These features can reflect the distribution and intensity of the frequency components in the motion data, and are often used to distinguish different types of actions.

Feature selection：The more features extracted, the higher the complexity of the classifier and the longer the computation time required, which affects the real-time performance of the recognition system. Therefore, feature selection is necessary to reduce the redundancy between different types of features. Commonly used feature selection methods include correlation-based feature selection, mutual information-based feature selection, and principal component analysis. These methods can effectively reduce the number of features while maintaining the accuracy of recognition.

Conclusion：In order to accurately recognize the actions of athletes, it is necessary to extract feature data from the original data collected. Time-domain features and frequency-domain features are two important types of features. Feature selection is necessary to reduce the complexity of the classifier and improve the speed of the recognition system. The selection of appropriate features can effectively improve the accuracy and efficiency of action recognition, which has important practical value for sports training and medical rehabilitation