基于多模态多标签学习的智能睡眠监测方法研究

睡眠障碍严重危害人民健康。多导睡眠图是睡眠障碍诊断的“金标准”，但测量负荷重，智能检测功能不完善，效率低，不适宜病人居家监护。家用睡眠监测的技术瓶颈在于：(1)减轻测量负荷，会造成部分体征信号的缺失，降低检测精度；(2)居家环境噪声的多样性和用户个体差异，严重影响模型的学习和泛化，降低检测稳定性；(3)千家万户的睡眠监测数据不断汇集到后台服务器，给计算资源和存储资源带来巨大的压力，降低了检测模型的更新效率。为了突破以上技术瓶颈，本项目拟：(1)研究多模态多标签数据的联合表征与推断方法，缓解模态缺失问题，提高检测精度；(2)研究多模态多标签模型的不确定性感知和自适应信息融合方法,增强检测稳定性；(3)研究多模态多标签增量学习方法，提高检测模型的更新效率。本项目旨在通过研究多模态多标签学习方法，推动家用睡眠监测技术的发展，并促进生物医学信号处理与机器学习理论的交叉融合。

Sleep disorder seriously endangers people's health. Polysomnography is the "gold standard" for the diagnosis of sleep disorders. However, it is not suitable for home monitoring because of its heavy measurement load, imperfect intelligent detection function and low efficiency. The technical bottlenecks of home sleep monitoring lie in: (1) reducing the measurement load will cause the loss of some physical signals and reduce the detection accuracies; (2) the diversity of noise in home environment and individual differences of users will seriously affect the learning and generalization of the model and reduce the detection stability; (3) the sleep monitoring data of thousands of households are continuously collected to the background server, which brings huge pressure on computing resources and storage resources, and reduces the efficiency of updating the detection model. In order to break through the above technical bottlenecks, the project plans to: (1) study the joint representation and inference method of multi-modal and multi-label data, alleviate the modal missing problems, and improve the detection accuracies; (2) study the uncertainty awareness and adaptive information fusion method of multi-modal and multi-label model, and enhance the detection stability; (3) study the incremental learning method of multi-modal and multi-label data to improve the update efficiency of detection model. The aim of this project is to advance the development of home sleep monitoring technology by studying multi-modal multi-label learning methods, and to facilitate the cross-integration of biomedical signal processing and machine learning theory.