基于Bi-LSTM的CSI手势识别算法  

CSI gesture recognition algorithm based on Bi-LSTM

作　　者：郭浩雨 冯秀芳[1] GUO Hao-yu;FENG Xiu-fang(College of Software,Taiyuan University of Technology,Jinzhong 030600,China)

机构地区：[1]太原理工大学软件学院,山西晋中030600

出　　处：《计算机工程与设计》2022年第9期2614-2621,共8页Computer Engineering and Design

基　　金：山西省自然科学基金项目(201801D121142)。

摘　　要：针对目前大多数基于CSI的手势识别方法存在精度偏低以及成本过高等缺陷,提出一种基于Bi-LSTM的CSI手势识别算法BGR (Bi-LSTM gesture recognition)。提取原始数据中的幅值信息,将其重构为长度一致的信号片段;利用PCA提取CSI信号的主成分特征,经过低通滤波除去背景噪声和多径效应干扰;在时域尺度上将连续手势动作信息输入到基于Bi-LSTM的特征融合模型中进行深层特征提取以及分类识别训练。在相关手势数据集上进行对比测试,其平均准确率达到91.49%。实验结果表明,该算法提高了手势识别的准确率,具有较强的适应性。Aiming at the disadvantages of most current CSI-based gesture recognition methods such as low precision and high cost,a Bi-LSTM based CSI gesture recognition algorithm BGR(Bi-LSTM gesture recognition)was proposed.The amplitude information in the original data was extracted and it was reconstructed into signal fragments of consistent length.The principal component analysis was used to extract the principal component characteristic information of the CSI signal,and the background noise and multipath effect interference were removed by low-pass filtering.The continuous gesture action information was inputted into the Bi-LSTM based feature fusion model on the time domain scale for deep feature extraction and classification recognition training.A comparative test was conducted on the relevant gesture data sets.The average accuracy reaches 91.49%.Experimental results show that the proposed algorithm improves the accuracy of gesture recognition and has strong adaptability.

关 键 词：信道状态信息 循环神经网络 手势识别 主成分分析 WI-FI 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]