Data fusion applied in near and mid infrared spectroscopy for crude oil classification. (15th May 2023)

• Record Type: Journal Article
• Authors: Moro, Mariana K.; de Castro, Eustáquio V.R.; Romão, Wanderson; Filgueiras, Paulo R.
• Is Part Of: Fuel. Volume 340(2023)
• Publication Date: 2023-05-15
• Subjects: Data Fusion -- PLS-DA -- Classification models -- Infrared spectroscopy -- Crude oils; Fuel -- Periodicals; Coal -- Periodicals; Coal; Fuel; Periodicals; 662.6
• DOI: 10.1016/j.fuel.2023.127580
• Languages: English
• ISSNs: 0016-2361

Abstract

Highlights: Data fusion strategies were applied to classify crude oil samples. PLS-DA models were built with fused Fourier-transform near and mid infrared spectra. Data fusion models showed high accuracy, reaching values equal to or greater than 94%. MIR and NIR spectroscopies could complement each other, improving the discrimination performance. Abstract: We report an application of a low-level data fusion for chemometric discrimination of crude oil samples by usual classifications, based on the combination of data obtained by means of the Fourier-transform near and mid infrared spectroscopy (NIR and MIR). The discriminant models were obtained by means of Partial Least Squares Discriminant Analysis (PLS-DA). The classification models based on individual spectra and on low-level fused spectra were then compared using their performance parameters. Data fusion achieved discrimination models with higher prediction accuracy for test samples, than the individual models, in the classification in poor or high-nitrogen content in crude oils and in the classification in light, medium and heavy crude oils. Data fusion also achieved a smaller prevision error rate and higher efficiency for test samples, for all classifications proposed in this study. The models built from data fusion showed high accuracy, reaching values equal to or greater than 94%. These results demonstrated that MIR and NIR spectroscopies could complement each other, through the data fusion strategy, yielding higher synergic effect and, consequently producing models with greater discriminant capacity.