基于24 GHz连续波多普勒雷达能量密度分布的非接触式睡眠动作检测
Non-Contact Sleep Motion Detection Based on Energy Density Distribution of 24 GHz Continuous Wave Doppler Radars
李佳程 1徐玉 1翁知翔 1唐震洲1
作者信息
- 1. 温州大学计算机与人工智能学院,浙江 温州 325035
- 折叠
摘要
睡眠动作是反映睡眠质量的重要生理指标.现有基于雷达的睡眠动作检测方法主要根据雷达信号原始能量的变化检测睡眠动作.由于不同目标动作幅度的不同会导致能量变化的差异,这些方法在检测不同目标的睡眠动作时准确率受限.为提高睡眠动作检测的准确率,提出了一种基于雷达能量密度分布的睡眠动作检测方法.首先,提出基于连续波多普勒雷达能量密度分布的检验统计量特征和显著性水平特征;随后基于上述特征引入XGBoost模型实现睡眠动作检测;最后,通过多种环境下的真实实验验证算法的有效性与鲁棒性.实验结果表明,所提出的检验统计量特征和显著性水平特征可以在降低训练样本需求的同时有效提高准确率,在多种环境下均能达到 95%以上的检测准确率.
Abstract
Sleep motion is an important physiological indicator reflecting sleep quality.The existing radar based sleep motion detection methods mainly detect the sleep motion according to the change of the original energy of radar signals.However,the accuracies of these methods are limited when detecting sleep motions of different targets,because different amplitudes of target motions may lead to differ-ences in energy changes.To improve the accuracy of sleep motion detection,a sleep motion detection method based on energy density distribution of radars is proposed.Firstly,the statistic and the significance level features based on radar energy density distribution are extracted.Then,the XGBoost model is introduced to achieve sleep motion detection.Finally,the effectiveness and robustness of the method are verified by real experiments in various environments.The experimental results show that the proposed statistic feature and significance level feature can effectively improve the detection accuracy while reducing the number of training samples,and can achieve more than 95%of detection accuracy in various environments.
关键词
连续波多普勒雷达/睡眠动作检测/密度分布/机器学习Key words
continuous wave Doppler radar/sleep motion detection/density distribution/machine learning引用本文复制引用
基金项目
浙江省自然科学基金面上项目(LY19F010010)
出版年
2024
国家科技期刊平台
NSTL
万方数据