Automatic preprocessing of laser-induced breakdown spectra using partial least squares regression and feed-forward artificial neural network: Applications to Earth and Mars data

• Published in: Spectrochimica acta. Part B: Atomic spectroscopy Vol. 171; p. 105930
• Main Authors: Ewusi-Annan, Ebo, Delapp, Dorothea M., Wiens, Roger C., Melikechi, Noureddine
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier B.V 01.09.2020; Elsevier BV
• Subjects: Analytical methods; Artificial neural network; Artificial neural networks; Automation; Background noise; ChemCam; Datasets; Geological samples; Laser induced breakdown spectroscopy; Lasers; Learning algorithms; Least squares method; LIBS; Machine learning; Mars; Neural networks; PLS; Prediction models; Preprocessing; Regression analysis; Response functions; Scale models; Spectra; Spectroscopy; Spectrum analysis; SuperCam; ChemCam; SuperCam; Artificial neural network; LIBS; PLS
• ISSN: 0584-8547; 1873-3565
• DOI: 10.1016/j.sab.2020.105930

Due to its relatively simple and versatile nature, laser-induced breakdown spectroscopy experiments can yield enormous amount of data that normally needs to be preprocessed to remove background signal, electron continuum, and noise, and for some applications, correct for the instrument response function and normalize the signal prior to conducting spectroscopic analysis. In experiments where the focus is on the analysis of samples of similar composition, preprocessing can be repetitive and tedious. We show that preprocessing of such LIBS data can be performed in an automated or semi-automated manner using machine learning tools. To demonstrate this approach, we apply partial least squares regression and artificial neural networks on two laser-induced breakdown spectra datasets. The first dataset is used to develop predictive models for abundances of various elements in geological samples analyzed by a laboratory model of ChemCam. The second dataset consists of spectra obtained from ChemCam as it interrogates Martian targets. We show that using the two machine learning techniques, we can predict the preprocessed spectra of samples with a relatively high accuracy for both datasets. [Display omitted] •Preprocessing of LIBS data acquired by Curiosity is performed using machine learning.•Partial least squares regression and artificial neural networks used to preprocess LIBS data.•Highly accurate predictions of preprocessed spectra acquired by Curiosity are obtained.