Research On The Intelligent Noninvasive Blood Glucose Detection System

With the improvement of living standards,the preference for high calorie food,sedentary and irregular work and rest and other urbanized living habits cause more and more II diabetics,which greatly endangers the national health.With the existing medical method,it’s impossible to cure the diabetes.It only can be controlled and alleviated to a certain extent.So diabetics need to be took blood glucose measurements several times one day.Traditional invasive and minimally invasive detection methods not only can bring psychological burden and physical agony,but also frequently detection can increase the risk of infection.Therefore,it is particularly important to develop a non-invasive blood glucose detection system.Based on the theory of nearinfrared spectroscopy,the calibration model between the photoplethysmography and the blood glucose concentration is established by using the algorithm of the extreme learning machine.The research content mainly consists of the following parts:1.Design the hardware acquisition system of photoplethysmography.The whole acquisition system is divided into front-end acquisition module,signal conditioning module and A/D conversion module,which respectively complete the functions of signal acquisition,signal filtering and amplification and analog-to-digital conversion.2.Filtering and denoising the photoplethysmography.As the photoplethysmography is a very weak physiological signal,it’s easily interfered by noise during the acquisition process.In this paper,the improved wavelet transform method is used to denoise the signal,effectively remove the noise such as power frequency interference,electromyography interference and baseline drift.And the threshold method,dynamic coefficient of variation method and autocorrelation function method are used to identify the signal interference section,improving the signal quality of the photoplethysmography.3.Establish the non-invasive blood glucose detection calibration model.In this paper,five pulse wave eigenvalues which have a strong relationship with blood glucose are selected to form the eigenvector matrix,and then the extreme learning machine is proposed to build the correction model,which is compared with multiple linear regression,partial least square regression and support vector regression.By comparing the correlation coefficient,prediction deviation and Clark grid error of the prediction results.The calibration model built by the extreme learning machine is proved the best when using the data collected in this paper.