An advanced image processing and multivariate statistical methodology to interpret Micro-EDXRF 2D maps

Abstract

Phosphogypsum (PG), a by-product of the fertilizer industry, is a potential source of rare earth elements (REEs) such as Lanthanum (La), Cerium (Ce), Neodymium (Nd), and Yttrium (Y). These elements were efficiently detected using micro-Energy Dispersive X-Ray Fluorescence (μ-EDXRF). Although a homogeneous REE distribution was expected in μ-EDXRF 2D maps, significant heterogeneity and variations in elemental associations (EA) were observed at a micrometric scale. To enhance and better interpret μ-EDXRF mapping results, a specialized image processing methodology was developed, incorporating Principal Component Analysis (PCA), Hierarchical Clustering (HC), and Multiple Linear Regression (MLA) which were applied to process and analyse 2D RGB pixel data. Identification of spatial overlaps, and multivariate correlations among the detected elements could be achieved. Notably, distinct EA patterns were found, with Ti, Ba, Y, and K playing a key role in REEs spatial distribution. Strong positive spatial correlations were identified between La and Ti, while Ce, Nd, and Y exhibited independent spatial distributions relative to La in certain sample areas. MLA further revealed strong EA between La, Ce, Nd, Y, and K, particularly in locations where Ti or Ba were also present. Additional elemental interactions were detected with Al, Cl, Ni, and Fe, with P and Cl showing significant correlations. Multicollinearity effects suggest strong interdependencies among elements. These findings highlight distinct REE spatial distributions within PG, demonstrating that mineralogical and compositional variations within the PG matrix influence REE spatial distribution patterns. Understanding these associations can improve strategies for REEs recovery from PG waste.