基于多模态异质信息融合的非接触式睡眠监测理论与关键技术研究

With the continuously improved requirements of modern sleep monitoring, non-contact sleep detection with single sensor cannot meet the requirements in both accuracy and performance. More comprehensive, precise and efficient theory and key technology for non-contact sleep monitoring have become one of the key issues in sleep medicine. The research on the theory and key technology of non-contact sleep monitoring based on multi-modal heterogeneous information fusion will be carried out in this project. This research mainly include the research on the sensing mechanism of multi-modal sleep physiological information, the extraction and screening of multi-modal sleep heterogeneity features, multi-modal sleep information fusion method, sleep quality assessment based on multi-modal heterogeneous information fusion and so on. In this research, multi-modal information fusion and machine learning are applied to the non-contact sleep monitoring technology. The advantages of combing the radio frequency and acoustic sensors are fully exploited. The "implicit" information of multi-modal sleep is deeply tapped. The multi-modal heterogeneous sleep information are complemented. The accuracy and robustness of non-contact sleep monitoring are improved. The project is the further study of non-contact sleep technology. It will explore new solutions to solve the problem of non-contact sleep monitoring. The research results will provide important research accumulation and technical support for the development of non-contact sleep monitoring in our country.

随着现代睡眠监测需求的不断提高，单一传感器进行非接触式睡眠监测在精度和性能上均不能满足相关需求，研究更加全面、精准、高效的非接触式睡眠监测理论与关键技术已成为睡眠医学重要问题之一。本项目将开展基于多模态异质信息融合的非接触式睡眠监测理论与关键技术的研究，研究内容主要包括多模态睡眠生理信息感知机理研究、多模态睡眠异质特征提取和筛选、多模态睡眠信息融合方法、基于多模态异质信息融合的睡眠质量评估等。研究中将多模态信息融合和机器学习应用到非接触式睡眠监测技术中，充分发挥射频和声传感器的联合优势，深度挖掘多模态睡眠“隐性”信息，实现多模态睡眠异质信息的互补，提高非接触式睡眠监测的准确性和鲁棒性。本项目是对非接触式睡眠技术的进一步深化研究，将为解决非接触式睡眠监测问题探索新的方案。研究成果将为我国非接触式睡眠监测领域的发展提供重要的研究积累和技术支撑。

随着现代睡眠监测需求的不断提高，单一传感器进行非接触式睡眠监测在精度和性能上均能满足相关需求，研究更加全面、精准、高效的非接触式睡眠监测理论与关键技术已成为睡眠医学重要问题之一。项目形成的多模态生理异质信息测量及融合技术突破了传统利用单一传感器进行生理信息测量的局限性，充分发挥射频、音频各自的优势，深度挖掘多模态生理“隐性”信息，实现多模态生理信息的互补，提高非接触式生理信息测量的完备性和鲁棒性。针对睡眠表层生理信息和睡眠呼吸障碍内在映射关系不清晰的问题，在国际上首次提出非接触式多模态睡眠感知技术，通过射频传感器、声传感器精确捕获睡眠生理信息，从多谱域多角度刻画呼吸、心跳和体动等信号，同时构建呼吸声、鼾声时频域、非线性特性和睡眠呼吸障碍的映射关系，最终实现了睡眠呼吸暂停、低通气等睡眠呼吸疾病和睡眠阶段的精准辨识。.本项目执行过程中，针对本领域缺乏睡眠呼吸障碍疾病患者数据支撑的问题，在“十二五”国家科技支撑计划首个基层监护监测新产品专项、“十三五”国家重点研发计划主动健康专项等一系列项目支撑下，与中南大学湘雅医院、江苏省人民医院、南京医科大学附属淮安医院、苏州大学附属第二医院等十余家知名三甲医院建立深度合作。截至目前，已在上述医院呼吸科、睡眠中心等科室积累10万+分钟、300+病例临床测试数据，突破大数据壁垒，初步建成此类传感器首个临床数据库。 围绕该工作累计发表 SCI 论文 24 篇，IEEE 系列论文 22 篇（Tran.期刊论文 8 篇），2020 年受邀撰写英文专著《Micro-Doppler Radar and its Applications》的第十章内容，2021 年受邀撰写英文专著《Contactless Vital Signs Monitoring》的第十四章内容。这部分研究工作已经引起国内外相关领域研究者的广泛关注，Google Scholar 总他引次数超过 350 次，先后得到 20 多位 IEEE Fellow 的正面引用。