体域网中多传感器数据融合算法的研究

发布时间：2018-01-05 21:31

本文关键词：体域网中多传感器数据融合算法的研究　出处：《吉林大学》2016年硕士论文　论文类型：学位论文

【摘要】：近年来,我国老龄人口的比例越来越高,其中独居老人占很大的比例。其健康问题已成为社会大众关注的一大焦点。近期发展迅速的无线体域网,为解决独居老人的健康问题提供了切实可行的解决方案。其中,基于可穿戴式节点的体域网更适用于老龄群体,能为其提供远程医疗监护服务。然而,体域网中某些关键技术如多源数据融合技术没有形成统一标准,复杂的生理信号如何更全面准确地表征人体状态等有待进一步的研究。为了能够让人体生理数据在保留人体特征信息的同时降低数据的复杂性,并通过综合分析这些数据获知人体的健康状态,本文对体域网下的多源传感器数据融合的相关算法进行了如下研究工作:(1)基于e-Health硬件平台搭建了体域网,并对数据的采集、预处理和传输进行了研究。由于该体域网难以获取人体异常状态下的生理数据,本文采用多参数智能重症监护数据库。该数据的预处理包括四个步骤:首先,对数据集的假警报数据进行分析和剔除;其次,对数据重采样处理,该过程使用FIR低通滤波器,进而避免了重采样时信号混叠;然后,对数据去噪处理,本文改进了小波阈值去噪法的阈值函数及阈值,从而达到更好的去噪效果;最后,对数据特征提取处理,本文用极大似然法处理离散信号,小波包变换法处理连续信号,这种分类提取信号特征的方法提高了信号的可靠性。(2)基于模糊神经网络解决了多项生理参数有效融合问题。本文首先介绍了多层前馈神经网络和模糊推理理论并分析了两者的不足之处。其次,详细分析了模糊神经网络的优点、结构及算法实现,最后,通过对比实验验证了模糊神经网络算法比单一算法对多源数据具有更好的融合效果。在此基础之上,本文构建了一种简捷、高效的模糊神经网络结构,并基于自适应理论提高了模型的学习速率,从而提高了模糊神经网络的性能和融合结果的准确性。本文在数据集的选取、识读及处理,数据融合的去噪、重采样、特征提取及模糊神经网络分类等方面的工作对于研究体域网中多传感器的数据融合有一定的借鉴意义。
[Abstract]:In recent years, the proportion of the elderly population in China is increasing, among which the elderly living alone account for a large proportion. Its health has become a major focus of public concern. Recently, the rapid development of wireless body area network. It provides a feasible solution to the health problems of the elderly living alone. Among them, the body-area network based on wearable nodes is more suitable for the elderly population and can provide telemedicine services for them. Some key technologies in body area network, such as multi-source data fusion, have not formed a unified standard. In order to make the physiological data of human body retain the characteristic information of human body and reduce the complexity of the data, how to represent the human body state more comprehensively and accurately by the complex physiological signal needs to be further studied. And through the comprehensive analysis of these data to know the state of human health. In this paper, the related algorithms of multi-source sensor data fusion under body domain network are studied as follows: 1) based on the e-Health hardware platform, the body area network is built and the data is collected. Preprocessing and transmission are studied. Because the body area network is difficult to obtain physiological data in abnormal state of human body, this paper adopts multi-parameter intelligent intensive care database. The preprocessing of this data includes four steps: first. Analysis and elimination of false alarm data in data sets; Secondly, the FIR low-pass filter is used in the process of data resampling, which avoids signal aliasing when resampling. Then, for data denoising, this paper improves the threshold function and threshold of wavelet threshold denoising method, so as to achieve a better denoising effect; Finally, for data feature extraction, this paper uses maximum likelihood method to process discrete signals, wavelet packet transform method to process continuous signals. This method improves the reliability of the signal. Based on fuzzy neural network, the problem of effective fusion of multiple physiological parameters is solved. Firstly, the multilayer feedforward neural network and fuzzy reasoning theory are introduced and their shortcomings are analyzed. The advantages, structure and algorithm implementation of fuzzy neural network are analyzed in detail. Finally, the comparison experiment shows that the fuzzy neural network algorithm has better fusion effect on multi-source data than the single algorithm. In this paper, a simple and efficient fuzzy neural network structure is constructed, and the learning rate of the model is improved based on adaptive theory. Thus, the performance of fuzzy neural network and the accuracy of fusion results are improved. In this paper, the data set selection, reading and processing, data fusion de-noising, resampling. The work of feature extraction and fuzzy neural network classification can be used for reference in the study of multi-sensor data fusion in body domain networks.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP202;TP212

【相似文献】