LIBS identification of lithology based on spectral lines of 7 main elements

Abstract

The laser induced breakdown spectroscopy combined with factor analysis and BP neural network is applied in lithology sorting and distinguishing for 9 standard samples of 5 types. Characteristic spectrum was formed by spectral lines of 7 elements which include Mg, Si, Al, Fe, Ca, Na and K, and were chosen according to the contents of the main elements in samples. The research object of each element was a selected peak of which the wavelength range was determined according to the shape and size. The Principal Component Analysis and Characteristic Spectrum Analysis are the dimensionality reduction method, this paper compared LIBS spectra analysis of them. Considering the samples are standard, under the "teachers" training mode of BP neural network is feasible. Using Principal Component Analysis (PCA) of factor analysis to analyse the full spectrum and characteristic spectrum. Then the full spectrum and the characteristic spectrum were respectively input to the BP neural network for lithology sorting and distinguishing. we did the same to the principal components of full spectrum and characteristic spectrum. Lithology sorting for samples by BP neural network in 4 cases, as follows: full spectrum, principal components of full spectrum, characteristic spectrum and principal components of characteristic spectrum. The highest identification accuracy of which is the characteristic spectrum reached 98.89%. Distinguishing for samples by BP neural network also in 4 cases, the achieved highest identification accuracy of which is the characteristic spectrum reached 98.89% too. The experimental results shown that the characteristic spectrum extracted from full spectrum can represent it for factor analysis and BP neural network analysis, and it was more accurate and efficient for sorting and distinguishing. According to the main elements in samples to extract characteristic spectrum from full spectrum can reduce amount of interference information in the full spectrum, which can improve the identification accuracy of BP neural network.

References

×
SciEngine
CART
CUSTOMER
中文
LOGIN
Search
SciEngine AI Intelligent Search