Prospection of pyrochlore and microlite mineral groups through Raman spectroscopy coupled with artificial neural networks

Niobium (Nb) and tantalum (Ta) concentrated in pyrochlore and microlite mineral groups, respectively, have attracted worldwide attention due to their importance to aerospace and electronics industries. This manuscript addresses the use of Raman spectroscopy coupled with artificial neural networks (ANNs) for improving the identification and characterization of mineral species belonging to pyrochlore and microlite mineral groups. Spectral data were collected in the 100–1400 cm−1 range and two baseline corrections, namely Asymmetric Least Squares (ALS) and Piecewise Linear Fitting (PLF) were performed and compared. In most cases, ALS achieved better performance in the removal of background noise with no elimination of important features of the original spectrum. The ANNs were fed with balanced datasets and based on different topologies with logistics, hyperbolic tangent, and rectified linear unit activation functions in the hidden layers. Pyrochlore and microlite minerals were identified by Raman spectroscopy. Multilayer Perceptron network classified minerals. Mineral classifier built from artificial neural networks and Raman spectra.