Classification Method And Applied Study For Spectral Data Of Laser-induced Breakdown Spectroscopy (LIBS)

Laser induced breakdown spectroscopy (LIBS) is a new type of element analysis technology and it bears the advantages of non-contact analysis, multiple elements simultaneous detection, real-time online analysis and so on. LIBS technique is a frontier analysis approaches in the field of spectral analysis, and there is a wide range of applications in petroleum, metallurgy, geology, environmental protection, military, aerospace and other fields. However, LIBS instruments are limited in introducing into China due to the expensive cost, the monopoly of core technology and even the significant strategic role. Therefore it provides significant theoretical and practical support to develop an innovative laser spectrum instrument with core independent intellectual copyrights, more complete functions and low costs.In this thesis, the spectra pre-treatment, qualitative and quantitative analysis, classification analysis were investigated based on the review for classification algorithm on the LIBS spectra data, which mainly includes optimization, implementation and integration of algorithm. The main research includes:(1)For the characteristic of LIBS spectra, robust principal component analysis(RPCA) algorithm was introduced to denoise and reduce the dimension of LIBS spectra, and then the support vector machine(SVM) was used for classification of9different kinds of steels on LIBS spectrum. The classification accuracy of steels was improved. And then the effectiveness and feasibility of the method was verified by comparison test;(2) For the instability of classification performance for single classifier, the combination classification model based on partial least squares (PLS) and SVM was generated by AdaBoost algorithm to classify9different kinds of steels on LIBS spectrum efficiently. It was verified that the proposed combination classification model can greatly improve the accuracy and stability of the classification model;(3)Spectra analysis module, the qualitative and quantitative analysis and pattern recognition module was designed and integrated into "The system of pre-treatment and analysis on LIBS spectra ". The research results can be applied to spectrum data classification and provide support of analyzing and processing data for the innovative laser spectrum instrument.