A Methodological Study Of LIBS Spectral Recognition Of Typical Rock Water Weathering

Rock water weathering is formed by a series of weathering effects on the dissolution,hydrolysis and hydration of rock minerals.The accurate identification of typical water weathering can provide direct evidence for the related rock water processes,and it is of great significance to possible water environment research of the planets and the earth resource exploration.Laser-induced breakdown spectroscopy(LIBS)uses a high-energy pulsed laser to ablate and excite experimental samples to a plasma state.The emission spectra of the plasma can be used to perform in-situ,rapid,multi-component,non-contact analysis of the elemental components in the sample.So combining laser-induced breakdown spectral with multiple classification and recognition algorithms can realize the classification of natural geological mineral samples,besides,it has important application value for planetary exploration and geological mineral field analysis.Based on the LIBS spectral experimental acquisition and spectral characteristics of natural rock water weathering and related mineral samples,this thesis focuses on the pretreatment,classification and decision fusion methods of LIBS spectral data,and seeks to establish an effective identification method of typical Rock water weathering.This thesis firstly introduces the LIBS spectrum acquisition experimental device,and it focuses on the non-uniformity characteristics of LIBS spectral of natural geological mineral samples,and pre-processing operations such as normalization,multi-spectral averaging,and wavelet denoising are performed to increase the stability of subsequent classification recognition.Besides,in order to improve the operational efficiency of the method model of classification and recognition,2 million multi-dimensional original LIBS spectral data is reduced to the 1614-dimensional characteristic spectrum on the basis of the element spectral line attribution.Then support vector machine and convolutional neural network method were applied to the classification of natural geological minerals respectively.Through parameter optimization,the above two methods respectively achieved recognition accuracy of 95.625%and 94.375%for 16 natural mineral samples.Based on the classical correlation analysis and combining with the element peak intensity ratio information,an improved correlation analysis method was proposed which achieved a recognition accuracy of 98.125%for 16 natural mineral samples.The improved self-organizing feature mapping neural network method model has a classification and recognition accuracy rate of 96.25%for 16 natural mineral samples combining the self-organizing feature mapping neural network method with traditional correlation analysis.The correlation results prove the validity of the classification methods and spectral preprocessing and spectral feature extraction.Due to the extremely complex composition and structure of natural mineral elements,there is a high degree of similarity between elemental components in similar mineral samples.So it is difficult for a single classifier to completely identify the sample correctly.In this thesis,the improved multi-classifier decision-level fusion method is used to combine the improved correlation analysis method,support vector machine and convolutional neural network.With the complementation of information between different classifiers,100%recognition accuracy was achieved for 16 natural mineral samples.The fusion method of the thesis proves that multi-classifier fusion can obtain the recognition results of natural geological minerals with better accuracy,robustness and stability,which has a guiding role in the application of LIBS technology in the field of mineral classification.