面向帕金森病的多模态异构协同感知方法

摘要

利用多模态异构传感器组成身体感知网络(body sensing networks),是连续感知用户日常行为的重要方法之一,但是能源消耗问题一直是限制其发展的主要原因.本文提出了一种面向帕金森病的多模态异构协同感知方法,以降低用户日常行为感知过程中的功耗.该方法将行为感知分为行为识别与状态监测,基于信息论确定识别或监测不同行为的最优传感器组合,进而利用一个多分类器建模的行为识别模型与多个二分类器建模的状态监测模型感知用户行为.通过在公开两个数据集上的实验可以看出,与传统的传感器全部持续工作的方法相比,该方法能够在保证对用户行为有效感知的同时,降低了数据传输和模型计算的功耗(MHEALTH上约40％,PAMAP2上约15％ ),从而延长感知网络的寿命,实现长时间持续的用户日常行为感知.%Using multimodal heterogeneous sensors to form body sensing networks (BSNs) is one of the most important ways to continuously sensing users' daily activities, but high energy consumption is the main reason for restricting its development. This paper presents a multimode heterogeneous collaborative sensing method for Parkinson's disease to reduce the energy consumption in sensing the daily activities by BSNs. The proposed method divides activity recognition into two sub-tasks which contain activity detection and status monitoring. And it uses one multi-classifier to model activity detection task and several binary classifiers to model status monitoring tasks, which are based on the chosen optimal sensor sets. Experimental results on two public dataset show that comparing with the conventional method whose sensors run all the time, the energy consumption on data transportation and model computation is reduced by 40％ in MHEALTH and 15％ in PAMPA2 approximately without losing activity-sensing accuracy. Thus it can help extend the lifetime of BSNs to sense users, daily activities long-termly and continuously.