A Novel XRF System for Cadmium Detection in Zinc Solutions Based on an Improved Baseline Correction Method and an RF-LSSVM Model

Cadmium is a typical by-product of zinc hydrometallurgy, which can compromise product quality and have significant environmental impacts if not properly handled. It has always been a challenge to find effective methods to measure its concentration due to the complex baseline and strong nonlinearity resulting from the background of high and varying zinc concentrations. In this article, we introduce a new detection method designed to measure Cd2+ concentration in zinc solutions with varying concentrations. Our study starts with a baseline correction method integrating adaptive smoothness penalized least-squares (asPLS) with negative-peak refinement steps, achieving improved baseline fitting. Subsequently, the random frog (RF) algorithm is employed for feature selection. Finally, a random frog-least-squares support vector machine (RF-LSSVM) model, trained on 112 samples from an X-ray fluorescence (XRF) spectrometer, addresses the issues of strong nonlinearity. Prediction results on the experimental testing set show the method’s effectiveness in measuring Cd2+ concentrations in zinc solutions with varying concentrations.