融合统计特征与CNN的WiFi人数识别方法
WiFi‑based people counting via fusion of statistical features and CNN
  
DOI:
中文关键词:  信道状态信息;人群计数;特征提取;卷积神经网络;门控机制
英文关键词:channel state information (CSI); crowd counting; feature extraction; convolutional neural network (CNN); gating mechanism
基金项目:国家自然科学基金(62172235)资助项目
作者单位
张载龙 南京邮电大学 物联网学院,江苏 南京 210003 
吴宇 南京邮电大学 物联网学院,江苏 南京 210003 
摘要点击次数: 183
全文下载次数: 258
中文摘要:
      为了提升多人场景下的人员数量识别精度,提出了一种基于信道状态信息(Channel State Information, CSI)统计特征(Statistical Feature, SF)与卷积特征的门控融合模型。首先,从原始 WiFi CSI中构建 12个与人员数量变化密切相关的统计特征指标,并通过系统性实验筛选出 6个最具判 别力的特征构成特征向量,同时利用卷积神经网络(Convolutional Neural Network, CNN)提取CSI的 空间特征。然后,使用门控机制将两类特征进行动态融合,实现了两类特征的协同利用。在 WiMANS 数据集上进行了实验,取得了 97.08% 的分类准确率,在 0~5人的不同人数下的识别率分 别为99.87%、99.65%、99.51%、95.22%、94.73%和93.51%,优于已有的分类模型。
英文摘要:
      To improve the accuracy of people counting in multi-person environments, this paper proposes a gated fusion model that combines statistical features (SF) and convolutional features extracted from channel state information (CSI). First, twelve statistical indicators closely related to variations in the number of people are extracted from raw WiFi CSI signals. Through systematic experiments, six of the most discriminative features are selected to construct a statistical feature vector. Meanwhile, spatial features of CSI are extracted using a convolutional neural network (CNN). A gated mechanism is then employed to dynamically fuse the two types of features, enabling collaborative feature utilization. Experiments conducted on the WiMANS dataset demonstrate that the proposed model achieves a classification accuracy of 97.08%, with recognition rates of 99.87%,99.65%,99.51%,95.22%,94.73%, and 93.51% for zero to five people, respectively, outperforming existing classification models.
查看全文  查看/发表评论   附件

你是第5855769访问者
版权所有《南京邮电大学学报(自然科学版)》编辑部
Tel:86-25-85866913 E-mail:xb@njupt.edu.cn
技术支持:本系统由北京勤云科技发展有限公司设计