Quantitative determination of heavy metals by CARS assisted extreme learning machine using x-ray fluorescence

To reduce environmental pollution, it is crucial to conduct research on efficient, inexpensive, and accurate ways to measure the concentration of heavy metals. However, currently used x-ray fluorescence (XRF) spectrometers are inaccurate in measuring the low concentration of heavy metals. To address the aforementioned issues, a new prediction framework was proposed and implemented in MATLAB, including denoising, background deduction, feature selection, and decision-making. First, the wavelet hard threshold denoising (WHTD) method and the iterative discrete wavelet transform (IDWT) method were applied to denoise and correct the baseline of spectrums, respectively. Then, a heavy metal estimation model based on competitive adaptive reweighted sampling assisted extreme learning machine (CARS-ELM) was proposed. In detail, the CARS algorithm was utilized to select characteristic spectral lines for the target element, and a non-linear relationship between the selected lines and element concentration was established using the ELM method. The results showed that the CARS-ELM model predicted the concentration of Cr, Cu, Zn, As, Cd, and Pb accurately, with R
²
values of 0.991, 0.999, 0.999, 0.994, 0.993, and 0.995, respectively. It is demonstrated that heavy metal concentration can be accurately predicted using the XRF technique in conjunction with the proposed framework.