【摘要】 针对人体动作识别问题，提出一种基于同步压缩短时傅里叶变换的人体动作识别方法。使用毫米波雷达进行人体动作数据的采集，将采集到的数据进行同步压缩短时傅里叶变换得到其时频图；然后使用卷积神经网络对不同动作进行微多普勒特征提取并分类。在数据采集部分，使用毫米波雷达进行数据采集，有效地避免了外界因素的影响；在时频分析部分，使用窗函数优化的同步压缩短时傅里叶变换提高了时频聚集性。实验结果表明，该人体动作识别系统对不同人体动作的识别率可达到91.7%。更多还原

【Abstract】 A method of human action recognition based on synchrosqueezing short-time Fourier transform(SSTFT) is proposed. The millimeter wave radar is adopted to collect human motion data,and then the data is synchronously compressed and short-time Fourier transform is used for it to obtain its time-frequency map. The convolutional neural network is used to extract the micro-Doppler features of different actions. In the data acquisition part,the use of millimeter-wave radarfor data acquisition can effectively avoid the influence of external factors. In the time-frequency analysis part,the SSTFT optimized with the window function improves the time-frequency aggregation. The experimental results show that the recognition rate of the human action recognition system for different human actions can reach 91.7%.更多还原