Data Fusion Computing Method Of Pulse Wave In Body Area Network

In recent years, with the bio-sensor technology and the rapid development in the Body Area Network, more and more wearable products used to monitor the human health are constantly entering people’s lives. Acquiring and recording the physiological health data such as pulse, respiration and body temperature have become realistic and feasible. However, physiological data in a continuous long period of time is too large. A large amount of data is not conducive to analyze, observe and extract valuable information of the human body. Therefore, the purpose of this study is to achieve a data fusion computing method. It can extract the concise and effective changes on human physiological state from the continuous and large amounts of human physiological data.In this dissertation, Body Area Network as the application environment of our research, pulse wave signal, breathing and body temperature as the research object and data fusion as the research method are applied to sense the daily health status with the equipment of Body Area Network.The main content of this dissertation includes four parts.(1) A pulse wave extraction algorithm based on the pulse wave of ascending convexity, named SPD algorithm, was proposed. With the soft thresholding and wavelet decomposition technique, SPD algorithm and effectively removes the high frequency glitch noise and low frequency baseline drift in pulse wave signal. After the first order differential treatment, the law of speed of the pulse wave amplitude changes is received. By envelope extraction process, the energy distribution in time-domain is obtained. The "base point" in the pulse wave signal is identified by the long tail of envelope. Because the ascending branch consisting of base points, systolic peaks and systolic valleys is caused by the physiological changes of arterial volume induced from left ventricular ejection, systolic valleys in the pulse wave signal are located by SPD method. The pulse wave can be extracted from the pulse wave signal according to the neighboring systolic valleys. Experimental results show that the proposed SPD algorithm has a high accuracy than the existing seven kinds of algorithms.(2) A pulse wave model based on Lognormal function to extract the physiology characteristic parameters was put forward. In this dissertation, the Lognormal function model is composed by four long tail unimodal pulse functions. A method of determining the parameters in the Lognormal function model by the plus or minus of the derivative and the zero position in second derivative of pulse wave is proposed. When the parameters has been determined, the fitting process of the sum of four Lognormal functions can be replaced with four times fitting process of a single Lognormal function. This approach is called segmented sequential local curve fitting method. The parameter vector containing 12 elements in the Lognormal function model are regarded as the physiological characteristics of pulse wave. The experimental results show that the proposed Lognormal function model is better than the traditional Gaussian function model in physiological meaning representation and waveform fitting precision.(3) A support vector machine classification method to identify the health status in daily life by the physiological characteristics of pulse wave was raised. The principal components analysis technology is applied to reduce the dimension of the physical characteristics of pulse wave. The physical characteristics after the dimension reduction is used as the input of the support vector machine to realize the accurate judging of human health state in daily life. Moreover, an improved particle swarm optimization algorithm is proposed to optimize the support vector machine in the penalty factor and Gaussian kernel function parameters. Based on the optimization, the support vector machine we designed in this dissertation can maximize the pulse wave under different health status. After examining by the more than 100000 pulse waves, the support vector machine, designed and implemented in this dissertation, can achieve an average precision ratio of 96.03%, an average recall rate of 100%, an average false positive rate of 2.63%, an average false negative rate of 5.43% and the average precision rate of 97% of health status identify in given conditions of awake static posture, high intensity exercise, drink and sleep.(4) A binary coding data fusion method in decision level was proposed. We obtain the health status not only form the physiological characteristics of the pulse wave but also from the physiological characteristics of temperature and respiration. When the health status derived form the three kinds of physiological signals are prepared, a new fusion result in decision level can be obtained with our fusion method. In the process of binary encoding fusion, the rank of fusion result in decision level is determined by our method of the stability of physiological signal. Furthermore, the fusion result in decision level is modeled by the weighted Markov chain. Experimental results show the prediction results using the weighted Markov chain model consistent with the actual results change trend, and the proposed method has more effective functions.The main purpose of our study is to get the human physiological health for Body Area Network device users in daily with data fusion method. Through computing, assessing and judging the physiological health state of device users in Body Area Network, the health perception and early warning can be realized. The health status of body can been sensed in real time. The methods proposed in this dissertation can provide support for the design and development of health perception device based on pulse wave in Body Area Network.