Sample Projection Space Algorithm Based On Spectral Recognition

Because of the unique characteristic of spectroscopy,such as simplicity of operation,short time of measurement and capability of recognizing different components simultaneously,it has been widely applied in industries.As for spectral analysis,Quantitative analysis is a common metthod.Qualitative analysis is used to classify the experimental subjects,such as partial least squares discriminant analysis(PLSDA),soft independent modeling(SIMCA)and so on.Although these algorithms have achieved good results,they require parameters optimization and parameters selection,etc.The thesis proposes a qualitative analysis method called the sample projection space algorithm.No need of parametric optimization,it shows faster recognition and less calculating costs.Three different practical spectra are combined to identify the samples.The main contents are as follows:1.The history and development of spectroscopic analysis technology has been introduced.Since the spectrum analysis has a relatively simple operation,and less damage to the samples,it has been widely applied in industries.Subsequently,three practical spectra used in this thesis are introduced,including their advantages and characteristics.2.This thesis simply introduces the princeple of thees algorithms,which are used in this thesis,i.e the partial least squares discriminant analysis(PLSDA),the soft independent modeling classification(SIMCA),random forest(RF),and the sample projection space(LRC).Although PLSDA,SIMCA and RF are all good classification algorithms,these algorithms have some disadvantages,such as requiring parameter optimization.However,the algorithm proposed in this thesis doesn't require parametric optimization.With simple linear recognition framework,it has faster recognition speed and need less calculation costs.3.This thesis introduces the samples used in the experiment,the software and toolbox for the data processing.The classification experiment of heavy metal contaminated mud clam and drug dosages is presented in this thesis.Compared with other algorithms,it presents that the algorithm proposed in this thesis is the best for three different spectral data experiments.Considering the LIBS spectra data can improve recognition rate,Kennard-stone divided sample algorithm and the self-organizing map(SOM)neural network clustering algorithm are combined with the proposed algrithom to improve the recognition rate of the LIBS spectra.