基于小样本学习的毫米波雷达手势识别方法  

Millimeter⁃Wave Radar Gesture Recognition Method Based on Small Samples Learning

作　　者：龚树凤[1] 李棋斌 施汉银 林超 李翔 GONG Shufeng;LI Qibin;SHI Hanyin;LIN Chao;LI Xiang(College of Information Engineering,Zhejiang University of Technology,Hangzhou Zhejiang 310023,China;China Comservice Huizhan Company,Hangzhou Zhejiang 311215,China)

机构地区：[1]浙江工业大学信息学院,浙江杭州310023 [2]中国通信服务浙江慧展科技公司,浙江杭州311215

出　　处：《传感技术学报》2023年第8期1235-1242,共8页Chinese Journal of Sensors and Actuators

基　　金：浙江省自然科学基金重点项目(LZ22F010005);浙江省教育厅科研项目(Y201839636);教育部产学合作协同育人项目(202102357003)。

摘　　要：针对大部分的手势识别方法在较小样本下可能出现过拟合和训练误差等问题而影响识别精度,提出了一种基于多通道调频连续波(FMCW)毫米波雷达在小样本场景下的手势识别方法。采用FMCW雷达获取前推、后拉、推拉、拉推、快速挥手、快速敲击等六类手势的回波信号,首先对各种手势的雷达回波信号进行预处理,然后利用二维傅里叶变换得到手势动作的距离-时间谱图和速度-时间谱图,最后基于度量学习设计了双流度量学习原型网络以提取手势动作的多维特征并实现手势识别。实验结果表明,所设计的双流网络能够在小样本条件下进行手势识别,准确率可达到98.3%。Most gesture recognition methods have problems such as overfitting and training error under the case of small samples,which may affect recognition accuracy.A gesture recognition algorithm is proposed based on multi⁃channel frequency modulated continuous wave(FMCW)millimeter wave radar in few⁃sample scenario.Six types of gestures such as forward push,backward pull,push⁃pull,pull⁃push,fast wave,and fast tap are designed according to the FMCW radar parameters and related settings,the distance and Doppler varia⁃tion over time information of these six types of gestures are then extracted by using Fourier transform and related algorithms and trans⁃formed into grayscale images for storage,and the extracted distance and time information is stored by using meta.The weighted fusion of the two types of features is achieved by using the prototype networks algorithm in meta⁃learning,and the multidimensional features of hand gesture motion are successfully extracted,which has high recognition accuracy of of 98.3%under the case of small samples.

关 键 词：毫米波雷达 手势识别 小样本学习 原型网络 

分 类 号：TN957.52[电子电信—信号与信息处理]