Research On Integrated Learning Modeling Methods For Blood Species Identification By Using Raman Spectroscopy

Raman spectroscopy technology has been widely used in drug identification,food safety testing,petrochemical analysis and other fields in recent years due to its advantages of non-destructive and high sensitivity.In terms of blood species identification,machine learning methods combined with Raman spectroscopy are more convenient and accurate than traditional analysis methods.In addition,the Raman spectrum data tends to be high-dimensional and there is more redundant information or noise interference,which affects the quality of the model.In this paper,the characteristics of strong anti-noise ability and high stability are borrowed from the ensemble learning method,the prediction accuracy of the blood species identification model is improved,and the characteristic wavelength selection is used to improve the model operation efficiency.The specific research contents are as follows:(1)Combining Random Forest algorithm and Ada Boost,the RF?Ada Boost blood species identification method is proposed.The Random Forest algorithm is used as the weak classifier of the Ada Boost integration framework.During the iterative process,the sample distribution and weights are continuously adaptively adjusted,and the model is trained into a strong classifier with good performance finally.Taking blood Raman spectrum data as an example,RF,SVM,ELM,KELM,BP and SAE as comparison methods,human and non-human blood species identification experiments are performed under the training sets of different scales.The experimental results show that the RF?Ada Boost method is able to get higher classification accuracy and stronger stability.(2)Combining BP neural network optimized by the Mind Evolutionary Algorithm and Bagging,the MEABP?Bagging blood species identification method is proposed.The Mind Evolutionary Algorithm is used to optimize the initial values of the BP neural network,shorten the network training time and avoid falling into local optimal values to affect the model quality,and then using the Bagging framework to bulid an integrated model to improve its stability and reduce generalization errors.Using MEABP,BP,SAE,SVM,ELM and KELM as comparative methods to conduct blood species bi-class and multi-class identification experiments,the results show that the identification accuracy and stability obtained by the MEABP?Bagging method is able to achieve best in all experimental models.(3)Combining Least Angle Regression and Successive Projections Algorithm,the LAR-SPA characteristic wavelength selection method for Raman spectrum is proposed.The Least Angle Regression algorithm and Successive Projections Algorithm are used to select the characteristic wavelengths of the Raman spectrum data in two stages to achieve the purpose of reducing the spectrum data dimension and data redundancy.Full wavelengths,CARS,UVE,LAR and SPA are used as comparison methods to verify the performance of the model on the drug and blood Raman spectroscopy datasets,and the results show that the LAR-SPA model can effectively implement characteristic wavelength selection and improve the modeling quality.