Pattern recognition algorithms applied to soil analysis with LIBS

JUAN VOROBIOFF; CARLOS A. RINALDI; BOGGIO ,NORBERTO; CHECOZZI,FEDERICO

Buenos Aires (On Line meeting)

Encuentro; 1st. International OnLine Meeting On LIBS; 2020

CNEA , UNSAM , U.Complutense de Madrid , U. de Lyon , UNAM (México) , U. de Zaragoza, U. de Bari (Italia) ,

Soil has an extremely complex and very diverse chemical composition, since it contains many components such as minerals, organic matter, living organisms, fossils, air and water . Advanced data processing from LIBS (Laser Induced Breakdown Spectroscopy) measurements are of great interest for soil analysis [2]. Signal analysis together with pattern recognition, derived from measurements made with LIBS techniques in conjunction with chemometric techniques, allows classifying and quantifying different analytes. However, the different settings of the LIBS measurements significantly affect the measured signals. In turn, there is a wide variety of chemometric techniques. All this generates a large volume of data and results that must be compared with each other, to determine the best methods to use. However, these methods are not definitive. When modifying the measurements, the chemometric techniques used must be modified, their internal parameters must be adjusted and / or the algorithm be changed. It should be noted that numerous data processing methods can be implemented. This processing of large volumes of data and its corresponding algorithm modification, if performed manually, is very tedious, time consuming and highly trained personnel. On the other hand, there are different chemometric analysis programs, as an example Origin Pro programs (Origin Lab Corporation, Northampton, MA, USA) [1], Microsoft Excel and Chemoface can be used. However, to obtain the final result, programs from different manufacturers must be combined with incompatible formats, and with many limitations and failures, which do not adequately adapt to the problem of interest. For this reason, proprietary software with advanced algorithms for data processing was implemented, which greatly simplifies processing tasks, improves the responses obtained and is easy to use for personnel not trained in data handling. This Pattern Recognition software, based on a graphical user interface (GUI) is friendly and simple to use. The Python programming environment was used to automate data reading and perform signal pre-processing, chemometric analysis, presentation of results, and method comparison. This interface contains a selection of signal processing algorithms. These algorithms are made with Pattern Recognition techniques based on Multivariate Data Analysis, Principal Component Analysis (PCA), Statistical Analysis, Discriminant Functions, Computational Intelligence and Neural Networks. As a result of this work, a simple to use measurement-oriented pattern recognition tool LIBS was obtained, which compares different data processing methods.