| 引用本文: | 屈乐乐，祝诗卉. 基于三通道CNN⁃GSAM⁃LSTFEM网络的雷达人体切向动作识别[J]. 雷达科学与技术, 2024, 22(2): 135-144.[点击复制] |
| QU Lele, ZHU Shihui. Radar Human Tangential Activity Recognition Based on Three⁃Channel CNN⁃GSAM⁃LSTFEM Network[J]. Radar Science and Technology, 2024, 22(2): 135-144.[点击复制] |
|
| |
|
|
| 本文已被：浏览 3159次 下载 469次 |
码上扫一扫！ |
|
|
| 基于三通道CNN⁃GSAM⁃LSTFEM网络的雷达人体切向动作识别 |
|
屈乐乐，祝诗卉
|
|
沈阳航空航天大学电子信息工程学院， 辽宁沈阳 110136
|
|
| 摘要: |
| 为了提高干涉雷达对人体切向动作的识别性能，本文提出一种基于三通道CNN?GSAM?LSTFEM网络的人体切向动作识别方法。首先利用一发二收的调频连续波（FMCW）雷达搭建干涉雷达平台采集人体切向动作回波数据，之后对每个接收通道的回波数据进行预处理，得到每个接收通道的多普勒时频图（DTFM）和双通道的干涉时频图（ITFM），然后将这3种时频图分别送入到3个并行的CNN?GSAM?LSTFEM网络进行训练，利用全局空间注意力模块（GSAM）和长短时特征提取模块（LSTFEM）增强卷积神经网络（CNN）的特征提取能力，最后将三通道提取的特征进行融合实现人体切向动作识别。实验结果表明，所提方法可有效提高人体切向动作的识别准确率，平均准确率高达98.77%。 |
| 关键词: 人体动作识别 干涉雷达 注意力机制 卷积神经网络 特征融合 |
| DOI：DOI:10.3969/j.issn.1672-2337.2024.02.003 |
| 分类号:TN958.95 |
| 基金项目:国家自然科学基金（No.61671310）；航空科学基金（No.2019ZC054004）；辽宁省兴辽英才计划项目基金（No.XLYC1907134）；辽宁省百千万人才工程项目基金（No.2018B21） |
|
| Radar Human Tangential Activity Recognition Based on Three⁃Channel CNN⁃GSAM⁃LSTFEM Network |
|
QU Lele, ZHU Shihui
|
|
College of Electronic Information Engineering, Shenyang Aerospace University, Shenyang 110136, China
|
| Abstract: |
| In order to improve the performance of interferometric radar for human tangential activity recognition, a human tangential activity recognition method based on three?channel CNN?GSAM?LSTFEM network is proposed in this paper. Firstly, an interferometric radar platform is constructed using a frequency modulated continuous wave （FMCW） radar with one transmitter and two receivers to collect the human tangential motion echo data. Subsequently, the echo data are preprocessed to obtain the Doppler time?frequency map （DTFM） for each receiving channel and the two?channel interferometric time?frequency map （ITFM）. Then, the three obtained time?frequency maps are separately fed into three parallel CNN?GSAM?LSTFEM networks for training. The global spatial attention module （GSAM） and long?short time feature extraction module （LSTFEM） are used to enhance the feature extraction ability of convolutional neural network （CNN）. Finally, the features extracted from the three channels are fused to achieve human tangential activity recognition. The experimental results show that the proposed method can effectively improve the recognition accuracy of human tangential activities and the average accuracy is as high as 98.77%. |
| Key words: human activity recognition interferometric radar attention mechanism CNN feature fusion |
|