A Study On Qualitative And Quantitative Analysis Method Based On LIBS Spectrum Data

Laser Induced Breakdown Spectroscopy(LIBS) has been widely recognized as a promising element analysis techniques since its inception with the advantages of remote control, real-time, non-contact, high sensitivity, short analysis time, etc, and has been widely used in many fields such as metallurgy, biomedical, cultural relics protection, materials, environment, military, space exploration and so on. With the rapid development of stoichiometry and data mining technology, a new and efficient method for qualitative and quantitative analysis aimed at developing full-featured, low-cost laser spectroscopy instrument has become an active research direction in the world. To carry out the LIBS product development with the aid of the intelligent information processing technology will bring significant theoretical and applied values.The research work of this paper is supported by the special funds of national instrument development. The research group has achieved some research results in spectral data preprocessing and spectral data analysis fields. According to the requirements of the project, the main goal of this paper is to explore new optimization method to achieve more accurate qualitative and quantitative analysis of spectral data, and develop the corresponding computer aided analysis system in order to provide a certain decision support for LIBS product development.The main research contents of this paper include:1?On the basis of the existing classification integration method and aiming at the problem that the classification accuracy of a single random forest base classifier is not high, the hybrid tree model of CART and C4.5 decision tree generation algorithm is proposed to improve the accuracy of classification. The simulation experiments show the feasibility and effectiveness of the proposed method.2?For the problem that the accuracy of the LIBS spectral data quantitative analysis using the partial least square method and the artificial neural network still needs to be improved, the support vector regression machine is introduced. At the same time, a new LSVR algorithm based on improved particle swarm optimization algorithm and support vector regression machine is proposed for the problem of parameter selection of support vector regression machine.The simulation experiment proves the validity and feasibility of LSVR.3?The corresponding software copyright is acquired by designing and implementing the LIBS/LIBRAS spectral preprocessing and analysis system.