| At present,there are more than 400 million people with diabetes in the world.The real-time monitoring of blood glucose level is of great significance for the treatment of diabetic patients.The traditional blood glucose monitoring method is mainly invasive monitoring method,which will bring pain to patients and fail to timely detect blood glucose,and it may lead to cross-infection of infectious diseases.Therefore,the research and application of new non-invasive blood glucose monitoring is expected to be high increasingly.In this paper,a non-invasive glucose detection technique based on photoacoustic effect is studied,and a multi-band fusion method is proposed to establish the glucose concentration prediction model.Firstly,in this paper,the mechanism of photoacoustic signal generation,propagation and detection is analyzed in detail on the basis of in-depth study of the basic principles of photoacoustic effects.It obtains that the relationship between the amplitude of photoacoustic signal and the optical and physical properties of medium.The spectrum of photoacoustic signal is related to laser modulation function and optical absorption function in medium space.Thus,the relationship between photoacoustic signal and glucose concentration is established.Secondly,in order to overcome the problem that single-band photoacoustic signal is susceptible to be interfered,this paper adopts partial least squares regression method and BP neural network algorithm to establish the mathematical model between the peak of multi-band photoacoustic signals and the concentration of glucose solution.Aiming at the limitation of partial least squares regression method in dealing with complex nonlinear problems,this paper proposes a variable screening method based on forward search method to build a glucose prediction model,so as to improve the performance of the model.Aiming at the problem of slow convergence speed and easy to fall into local extremum of BP neural network,the activation function,learning rate and weight update were improved to train a better glucose prediction model.Finally,a photoacoustic noninvasive blood glucose detection system is built,which is mainly composed of the hardware part of photoacoustic signal acquisition and the software part of data processing.The spectrum information of photoacoustic signal of 0～ 500 mg / dl glucose solution is collected by Lab VIEW control spectrum analyzer.It is verified that the amplitude of photoacoustic signal and the number of spectrum peak points increase with the increase of glucose concentration.The prediction model of glucose concentration in three frequency bands is establish by partial least squares regression with variable selection method,and the predicted root mean square error of prediction is 20.95mg/dl.The BP neural network establish the glucose concentration prediction model of five frequency bands through multiple training,and the predicted root mean square error is 15.25mg/dl.The results show that the accuracy of glucose concentration prediction for multiple frequency band photoacoustic signal fusion was feasible. |
