基于小波分析的手势无线感知方法  

Wireless Gesture Sensing Based on Wavelet Analysis

作　　者：魏玉婷 王森 逄德明 陈颖文 WEI Yuting;WANG Sen;PANG Deming;CHEN Yingwen(School of Computer,National University of Defense Technology,Changsha 410015,China;National Informatics Centre,Beijing 100038,China)

机构地区：[1]国防科技大学计算机学院,湖南长沙410015 [2]国家信息中心,北京100038 [3]国防科技大学计算机学院计算科学系,湖南长沙410015

出　　处：《移动通信》2022年第7期57-61,共5页Mobile Communications

摘　　要：无线技术的快速发展,使得基于无线信道状态信息来进行无设备人类活动的感知成为可能。基于连续小波变换和迁移学习的方法采用智能无线感知原理,重点研究了应用在手势识别场景下的方法和实现。首先将收发器收集到的原始信道状态信息进行数据预处理,去除载波频率偏移和采样频率偏移以及噪声;其次,介绍了连续小波变换的原理,并使用其生成谱图;最后,使用预训练好的卷积神经网络进行手势分类,平均识别准确率在94%左右。With the rapid development of wireless technology, device-free human activity sensing based on wireless channel state information has become possible. Based on continuous wavelet transform and transfer learning, according to the principle of intelligent wireless sensing, the method and implementation of gesture recognition are investigated. Firstly, the original channel state information collected by the transceiver is preprocessed to remove carrier frequency offset, sampling frequency offset, and noise. Secondly, the principle of continuous wavelet transform is introduced and used to generate spectrograms. Finally, utilizing the pre-trained convolutional neural network for gesture classification, the average recognition accuracy is around 94%.

关 键 词：智能无线感知 手势识别 连续小波变换 图像识别 

分 类 号：TN929.5[电子电信—通信与信息系统]