Measurement of potentially toxic elements in the soil through NIR, MIR, and XRF spectral data fusion

Abstract This paper aims to investigate the feasibility combination of X-ray fluorescence (XRF), near-infrared (NIR), and mid-infrared (MIR) sensors for the detection of seven key monitoring elements in the soil. Two strategies for data fusion were adopted: (i) the XRF characteristic bands were fused with all spectral data of NIR and MIR separately, and (ii) the XRF characteristic bands were fused with the characteristic bands of NIR and MIR, respectively. Also, different feature extraction methods were compared. The best feature extraction methods for XRF-NIR and XRF-MIR models were principal component analysis (PCA) and successive projections algorithm (SPA). The modeling results showed that strategy (ii) showed better performance, and the XRF-MIR model provided more accurate results than the XRF-NIR model. Both XRF-MIR and XRF-NIR methods improved the accuracy of As, Cr, Cu, Ni, and Zn. The XRF-MIR obtained the best predictions, and the determination coefficients (R2) were 0.93, 0.98, 0.98, 0.95, and 0.98. For Pb and Cd, the measurement could be obtained by XRF alone, and the corresponding R2 were both 0.98. The results confirmed that sensor fusion can effectively improve the accuracy of the spectrometer in detecting metal elements in the soil.