[1]陶 华,张向群,等.基于轻量化 GSSTANet 毫米波雷达三维点云手势识别方法[J].机械与电子,2026,44(06):71-80.
　TAO Hua,ZHANG Xiangqun,et al.A Lightweight GSSTANet Method for Millimeter-wave Radar 3D Point Cloud Gesture Recognition[J].Machinery & Electronics,2026,44(06):71-80.
点击复制

基于轻量化 GSSTANet 毫米波雷达三维点云手势识别方法()
分享到：

《机械与电子》[ISSN:1001-2257/CN:52-1052/TH]

卷:
44
期数:
2026年06期
页码:
71-80
栏目:
智能检测
出版日期:
2026-06-27

文章信息/Info

Title:
A Lightweight GSSTANet Method for Millimeter-wave Radar 3D Point Cloud Gesture Recognition
文章编号:
1001-2257 ( 2026 ) 06-0071-10
作者:
陶 华 1 ; 2 ; 张向群 1 ;  2 ; 杜根远 1 ;  2 ; 张世功 3
1. 华北水利水电大学信息工程学院,河南 郑州 450011 ;?
2. 许昌学院信息工程学院,河南 许昌 461000 ;
3. 贵州理工学院理学院,贵州 贵阳 550025
Author(s):
TAO Hua1 ;  2 ;  ZHANG Xiangqun1 ;  2 ;  DU Genyuan1 ;  2 ;  ZHANG Shigong3
( 1.School of Information Engineering , North China University of Water Resources and Electric Power , Zhengzhou 450011 , China ;
?2.School of Information Engineering , Xuchang University , Xuchang 461000 , China ;
3.College of Science , Guizhou Institute of Technology , Guiyang 550025 , China )
关键词:
手势识别; 轻量化网络; 图神经网络; 星操作; 毫米波雷达
Keywords:
hand gesture recognition ;  lightweight network ;  graph neural network ;  star operation ;  millimeter-wave radar
分类号:
TP183 ;TN957.52
文献标志码:
A
摘要:
针对现有基于深度卷积神经网络的手势识别方法模型参数量大、难以满足资源受限嵌入式设备轻量化需求的问题,基于 mmEgoHand 提供的预估计骨骼点序列,设计一种轻量化图神经网络( GSSTANet )。该网络借鉴 GTNet 的全局注意力、 StarNet 的高维非线性映射特性以及融入时空交替处理思想,构建高效的手势骨骼识别模型。具体而言,引入基于图卷积与星注意力的空间建模模块,有效捕捉手部骨骼点间的复杂结构关系;采用星操作实现隐式高维特征映射,在不增加网络宽度下显著提升表达能力;通过连续时空交替机制,分别建模空间依赖与时间动态后深度融合,实现充分的时空信息提取。实验结果表明,所提方法在 mmVR 数据集上准确率达 94.22% ,模型参数量仅 0.37×106 ,模型大小压缩至 1.43 MB 。所提方法为轻量化、可部署的毫米波雷达三维点云手势识别提供了新思路。
Abstract:
To address the challenge of large model parameters in existing deep convolutional neural network based hand gesture recognition methods , which hinders their deployment on resource constrained embedded devices , we propose a lightweight graph neural network named Graph Star Spatial Temporal Alternating Net ( GSSTANet ) based on pre-estimated skeleton point sequences provided by mmEgoHand.The proposed network integrates the global attention mechanism of GTNet ( Graph Transformer ), the high-dimensional nonlinear mapping characteristics of StarNet , and a spatial temporal alternation to construct an efficient hand gesture recognition model for skeletal data.Specifically , a spatial modeling module based on graph convolution and star attention is introduced to effectively capture the complex structural relationships among hand skeletal points.The star operation ( element wise multiplication ) is employed to achieve implicit high dimensional feature mapping , significantly enhancing representational capacity without increasing network width.Furthermore , a sequential spatial temporal alternation mechanism is adopted , spatial dependencies and temporal dynamics are modeled separately and then deeply fused , enabling comprehensive extraction of spatio temporal information.Experimental results demonstrate that the proposed method achieves an accuracy of 94.22% on the mmVR dataset , with only 0.37×106 parameters and a model size of 1.43 MB.This work provides a new approach for lightweight and deployable hand gesture recognition using millimeter-wave radar 3D point clouds.

参考文献/References:

[ 1 ] Kuo P H , Shen Y C , Feng P H , et al.Transfer-learning-based gesture and pose recognition system for human-robot interaction : an Internet of things application [ J ] .IEEE Internet of Things Journal , 2024 , 11 ( 21 ): 35376-35389.

[ 2 ] Shi Yuhang , Qiao Lihong , Shu Yucheng , et al.Semi supervised FMCW radar hand gesture recognition via pseudo-label consistency learning [ J ] .Remote Sensing , 2024 , 16 ( 13 ): 2267.
[ 3 ] Mamlook R E A , Aljohani A.Advancing arabic sign language recognition : dataset validation [ J ] .Arabian Journal for Science and Engineering , 2025 , 50 : 20163-20183.
[ 4 ] Prakash K S , Kunju N.An optimized electrode configuration for wrist wearable emg-based hand gesture recognition using machine learning [ J ] .Expert Systems with Applications , 2025 , 274 : 127040.
[ 5 ] Wang Fei , Gao Yiao , Lan Bo , et al.U-shape networks are unified backbones for human action understanding from WiFi signals [ J ] .IEEE Internet of Things Journal , 2024 , 11 ( 6 ): 10020-10030.
[ 6 ] Gao Ruiyang , Li Wenwei , Liu Jinyi , et al.Wicgesture : meta-motion-based continuous gesture recognition with WiFi [ J ] .IEEE Internet of Things Journal , 2023 , 11 ( 9 ): 15087-15099.
[ 7 ] Li Mengning , Wang Wenye.Hybrid zone : bridging acoustic and WiFi for enhanced gesture recognition [ C ] ∥ IEEE INFOCOM 2024 IEEE Conference on Computer Communications.New York : IEEE , 2024 : 981-990.
[ 8 ] Wang Zhengjie , Liu Fei , Li Xue , et al.A survey of hand gesture recognition based on FMCW radar [ C ] ∥ ICCIP ’ 22 : Proceedings of the 8th International Conference on Communication and Information Processing , 2023 : 73-79.
[ 9 ] Tang Gaopeng , Wu Tongning , Li Congsheng.Dynamic gesture recognition based on FMCW millimeter wave radar : review of methodologies and results [ J ] .Sensors , 2023 , 23 ( 17 ): 7478.
[ 10 ] Hao Zhanjun , Sun Zhizhou , Li Fenfang , et al.Millimeter wave gesture recognition using multi-feature fusion and transformer [ J ] .Scientific Reports , 2024 , 14 : 13758.
[ 11 ] Yan Baiju , Wang Peng , Du Lidong , et al.mmGesture : semi-supervised gesture recognition system using mmWave radar [ J ] .Expert Systems with Applications , 2023 , 213 : 119042.
[ 12 ] Zhang Xianqun , Ge Zhizhou , Lu Kai , et al.Hand gesture recognition method based on mmWave radar with MobileViT and knowledge distillation [ C ] ∥ SPIE International Conference on Signal Processing Systems , 2024.
[ 13 ] Zhang Huafeng , Liu Kang , Zhang Yuanhui , et al.TRANS CNN-based gesture recognition for mmWave radar [ J ] .Sensors , 2024 , 24 ( 6 ): 1800.
[ 14 ] Chen Honghong , Wang Xiangyu , Hao Zhanjun , et al. Rageosense for smart home gesture recognition using sparse millimeter wave radar point clouds [ J ] .Scientific Reports , 2025 , 15 : 15267.
[ 15 ] Li Yang , Liu Yi , Li Hao , et al.Human pose estimation using millimeter wave radar based on transformer and pointnet + + [ J ] .Computer Science , 2025 , 52( 6A ): 240400169.
[ 16 ] Li Wenwei , Liu Ruofeng , Wang Shuai , et al.Egocentric human pose estimation using head-mounted mmWave radar [ C ] ∥ SenSys ’ 23 : Proceedings of the 21st ACM Conference on Embedded Networked Sensor Systems , 2024 : 431-444.
[ 17 ] Lv Yizhen , Zhang Tingting , Wang Zhijian , et al.mmEgoHand : egocentric hand pose estimation and gesture recognition with head-mounted millimeter-wave radar and IMU : version 2 [ PP / OL ] .V2.arXiv ( 2025-07-06 )[ 2026-02-10 ] .https : ∥arxiv.org / abs / 2501.13805.
[ 18 ] Zhou Wei , Wang Qian , Jin Weiwei , et al.Graph transformer for 3D point clouds classification and semantic segmentation [ J ] .Computers and Graphics , 2024 , 124 : 104050.
[ 19 ] Ma Xu , Dai Xiyang , Bai Yue , et al.Rewrite the stars [ C ] ∥2024 IEEE / CVF Conference on Computer Vision and Pattern Recognition ( CVPR ) .New York :IEEE , 2024 : 5694-5703.
[ 20 ] Yan Sijie , Xiong Yuanjun , Lin Dahua.Spatial temporal graph convolutional networks for skeleton-based action recognition [ C ] ∥Proceedings of the AAAI Conference on Artificial Intelligence , 2018 , 32 ( 1 ): 7444-7452.
[ 21 ] Liu Ziyu , Zhang Hongwen , Chen Zhenghao , et al.Disentangling and unifying graph convolutions for skeleton-based action recognition [ C ] ∥2020 IEEE / CVF Conference on Computer Vision and Pattern Recognition ( CVPR ) .New York : IEEE , 2020 : 143-152.
[ 22 ] Chen Yuxin , Zhang Ziqi , Yuan Chunfeng , et al.Channel-wise topology refinement graph convolution for skeleton-based action recognition [ C ] ∥2021 IEEE / CVF International Conference on Computer Vision ( ICCV ) .New York : IEEE , 2021 : 13359-13368.
[ 23 ] Wang Yue , Sun Yongbin , Liu Ziwei , et al.Dynamic graph CNN for learning on point clouds [ J ] .ACM Transactions on Graphics , 2019 , 38 ( 5 ): 146.

相似文献/References:

[1]许金鹏,张亚,吕铁钢,等.基于MATLAB的静态手势分割与识别研究[J].机械与电子,2017,(09):73.[doi:1001-2257(2017)09-0073-04]
　XU Jinpeng,ZHANG Ya,LYU Tiegang,et al.Research on Static Gesture Segmentation and Recognition Based on MATLAB[J].Machinery & Electronics,2017,(06):73.[doi:1001-2257(2017)09-0073-04]
[2]范 涛,王明泉,张俊生,等.基于轻量化 YOLOv4 的轮毂内部缺陷检测算法[J].机械与电子,2023,41(02):3.
　FAN Tao,WANG Mingquan,ZHANG Junsheng,et al.Internal Defect Detection Algorithm of Wheel Hub Based on Lightweight YOLOv4[J].Machinery & Electronics,2023,41(06):3.

备注/Memo

备注/Memo:
收稿日期: 2026-04-17
基金项目:河南省科技厅科学技术研究项目( 242102210067 , 252102521071 );河南省重点研发专项( 241111212500 )
作者简介:陶 华 ( 2000- ),男,河南焦作人,硕士研究生,研究方向为毫米波雷达手势识别;杜根远 ( 1974- ),男,河南许昌人,博士,教授,硕士研究生导师,研究方向为雷达探测与成像,通信作者, E-mail : dugy@xcu.edu.cn ;张向群 ( 1978- ),女,河南许昌人,博士,教授,硕士研究生导师,研究方向为雷达探测与成像。
更新日期/Last Update: 2026-08-26