Study Of Laser Induced Breakdown Spectroscopy Quantitative Analysis Methods Based On Machine Learning

Laser induced breakdown spectroscopy is an atomic emission spectroscopy technique,which has great application potential in material composition analysis.As a potential detection technology,LIBS has many advantages: no sample preparation,fast real-time analysis,nondestructive testing,remote analysis,multi-element simultaneous determination and so on.Therefore,the technology has been used in many fields,such as environmental monitoring,antiquities identification,food safety,geological exploration and so on.LIBS quantitative analysis has always been a hot and difficult point.This paper mainly studies the LIBS quantitative analysis method based on machine learning.The main contents are as follows:(1)To introduce the research background and significance of the subject,the principle,development course and research status of LIBS technology,summarize the application of machine learning methods in LIBS analysis,and the importance of variable selection in LIBS analysis.(2)The construction of LIBS analysis system,optimization of experimental parameters,sample preparation,raw spectral data acquisition and outlier screening are introduced.(3)The noise and baseline in the original LIBS signal are analyzed and discussed,and the discrete wavelet transform is used to reduce noise and baseline removal.The variable selection in LIBS is discussed,and the difference between the classical variable selection method PLS-VIP and the "two-stage variable selection method" proposed in this paper is compared.the result shows that the proposed variable selection method is more efficient and more universal than the first one.(4)Univariate analysis method and machine learning methods were used to establish calibration models respectively,the result shows that the use of machine learning methods can improve the ability of LIBS quantitative analysis.Taking the variable selection result S3 of the PLS-VIP method and the variable selection result S2 of the "two-stage variable selection method" as the input of the five machine learning methods,and comparing the performance of the machine learning models,the result shows that the machine learning models with S2 as input is generally better than the machine learning models with S3 as input.(5)The quantitative analysis algorithm library is established based on the noise reduction,baseline removal method,variable selection method and various machine learning methods used LIBS this paper.The LIBS quantitative analysis module is integrated on the LIBS analysis software based on the Qt platform.To sum up,the article based on the LIBS quantitative analysis,the noise and baseline in the original spectrum were corrected by discrete wavelet transform,the effective variables were extracted from the spectrum by two-stage variable selection method,and the LIBS quantitative analysis models were established by machine learning methods.The experimental results are of great significance to improve the precision and accuracy of LIBS quantitative analysis and promote the engineering application of LIBS.