Decoupling Free-Iron-Oxide Masking Enables Hyperspectral Quantification of Microplastics in Iron-Rich Purple Soil

Reliable hyperspectral measurement of microplastics in soil is compromised by structured mineral interference. Here, we identify free iron oxides (Fed) as a directional, gradient-dependent masking factor in iron-rich purple soil and develop target-orthogonal, gradient-weighted external parameter orthogonalization (TO-GW-EPO) to remove iron-associated spectral variation while preserving polymer information. A 5 × 5 factorial design covered measured Fed contents of 0.08-7.84 wt % and polyethylene (PE) loadings of 0-5 wt %. Increasing Fed depressed the reflectance continuum, attenuated C-H-related PE features near 1200 and 1395 nm, and weakened the concentration-peak-depth relationship; the 1395 nm masking index reached 79.0%, and r(PE,D1395) decreased from 0.94 to 0.13. Relative to standard EPO, TO-GW-EPO increased target-signal retention from 65.0% to 82.5%. Coupled with competitive adaptive reweighted sampling and a one-dimensional convolutional neural network, the workflow achieved Rp2 = 0.89, RMSEP = 0.63 wt %, RPD = 2.98, and an estimated LOD of 0.95 wt % for PE. These results support rapid screening and semiquantitative-to-quantitative analysis of relatively high-load samples under iron-rich mineral backgrounds. Extending accurate prediction to lower environmental concentrations remains a priority for future optimization. PET-specific reconstruction and retraining achieved Rp2 = 0.78 and RPD = 2.12, supporting workflow-level transferability to PET. This study establishes a physically constrained and interpretable strategy for reliable hyperspectral measurement under structured mineral-matrix interference.