Spectrum Analysis Of Blood Based On Statistical Machine Learning

In recent years,artificial intelligence technology based on statistical machine learning has developed into a powerful tool in many fields.With the rapid development of spectral technology,there is no doubt that using advanced statistical machine learning methods to predict and analyze spectral data will become a new direction in the field of spectral analysis.In this paper,we mainly discuss the application of statistical machine learning in the field of spectral analysis from two aspects: species identification of animal blood by fluorescence spectrum in vitro and non-invasive spectrum detection of blood glucose.We are committed to research more accurate and efficient methods of recognition and classification of unknown blood samples and prediction of blood glucose concentration.In the work of animal blood fluorescence spectrum recognition and classification,aiming at the collected fluorescence spectrum data,this paper studies the species recognition method based on traditional machine learning and DBN.In the research process,firstly,wavelet threshold denoising is used to preprocess the spectral data and remove the noise of the spectral data.Then,for the traditional machine learning classification model,PCA is used to extract the features of the spectral data.Finally,LR,DT,GBDT and DBN models are constructed to identify and classify,and the cross validation method is used to evaluate the model.In addition,this paper proposes a new method to identify and classify the fluorescence spectrum data of animal blood in vitro.In this method,DBN is used as the feature extractor to extract 200 depth features of the spectrum data of each blood sample.Then the features selected by the discriminant statistics are used as the input of the model for recognition and re-classification,that is,the recognition and classification model of the discriminant statistics "+" depth features "+" traditional machine learning method.The results show that compared with the single traditional machine learning classification model and DBN classification model,both the recognition efficiency and the recognition accuracy have been greatly improved.Therefore,DBN can effectively extract the features and achieve better classification.This method has great potential in forensic investigation,wildlife investigation and import and export detection.In the work of non-invasive blood glucose spectrum detection,the method of blood glucose concentration prediction based on SVR is studied for the collected near-infrared spectrum data.In the process of research,firstly,considering the influence of human background information and noise on spectral data,a preprocessing method of spectral data based on position average and wavelet threshold denoising is proposed,which can effectively eliminate background information and spectral noise.Then the SVR prediction model was established for different subjects to predict the blood glucose concentration.Theresults show that the accuracy of SVR model is better than the traditional model.Among them,the goodness of fit of the two models is 0.95 and 0.947 respectively,and the percentage of the predicted value distribution in the area a of Clark error grid is 98.82% and97.65%,respectively.This result has a high clinical validity,which lays a foundation for the further application of statistical machine learning in noninvasive blood glucose spectrum detection.