| Diabetes has gradually become a common disease with the improvement of people’s living standards,and blood glucose testing plays an essential role in both the prevention and treatment of diabetes.However,most of the existing blood glucose measurement methods may cause mental and physiological harm to the human body or have more stringent measurement requirements with higher testing costs.In the face of increasingly frequent blood glucose measurements,there is an urgent need for an accurate,convenient and non-invasive testing method to alleviate patients’mental and financial stress.Raman spectroscopy can be applied to blood glucose testing because of its strong ability to identify specific chemical bonds as well as its weak Raman scattering of water.To address the above situation,this paper will investigate the algorithms related to Raman spectroscopy signal processing in the case of blood glucose detection.The main contents are as follows.(1)A blood glucose measurement experimental platform based on spatial heterodyne Raman spectrometer was built.With the advantages of high resolution and system stability,the spatial heterodyne Raman spectroscopy technique is more suitable for direct blood glucose measurement.The excitation light source was 830 nm,which had a good inhibitory effect on fluorescence,as well as the short scanning time and high safety of measurement power.(2)The Raman spectral data of blood was simulated and used to compare multiple pre-processing algorithms.In the processing of Raman spectroscopy,a variety of noises are introduced depending on the detection equipment,environment,and the object to be measured-especially in blood glucose detection,where the glucose molecule signal is weak,and the noise needs to be removed by pre-processing.After comparing the evaluation indicators of various preprocessing algorithms,it is concluded that in the denoising algorithm,after adding 15d B white noise,the complete set of adaptive noise empirical mode decomposition algorithm has the highest signal-to-noise ratio(SNR=34.5522)and the lowest mean squared error(MSE=0.0200).Meanwhile,the least squares method has the smallest root mean square error(RMSE=0.7314)among the baseline removal methods.(3)An 1D-Res Net structure based on residual network was established for quantitative analysis of Raman spectral data.The glucose concentration in blood is low,while conventional modeling method is unable to achieve precise caculations of blood glucose based on Raman spectral data due to its low accuracy.In this paper,we improved the network structure based on residual network by combining the one-dimensional characteristics of Raman spectral data,so that the network has good computational accuracy in processing one-dimensional Raman spectral signals.The algorithm had the best performance in the comparison of partial least squares regression,support vector regression and Res Net in the glucose solution experiment.The regression coefficient(R~2)reached 0.8627 with a root mean square error(RMSE)of 0.8393.(4)A detection method based on the interaction between glucose and specific substances was proposed to enhance the detection effect.It is difficult to directly detect glucose molecules in the blood,however,glucose can react specifically with some substances,and the concentration of glucose can be detected by detecting the reactants.In this paper,based on glucose molecules’characteristics of easily reversible reaction with phenylboronic acid in a weak base environment,we prepared4,4’-biphenyldiboronic acid for the reaction.The regression coefficient of the produced complex was increased by 11.6%under 1D-Res Net,indicating that this method could effectively improve the detection accuracy of glucose molecules. |
PDF Full Text Request