| Abstract Scope |
Electronic waste (e-waste) presents a growing sustainability challenge due to its complex composition of rare earth elements, hazardous organics, and non-recyclable plastics. This study introduces a confidence-aware classification pipeline that combines mid-infrared hyperspectral imaging (HSI), spectral angle mapping (SAM), and iterative machine learning to enable pixel-level material identification. A curated spectral library spanning artificial materials, minerals, and organics was used to generate pseudo-labels with confidence scores derived from SAM similarity. High-confidence samples from seven consumer electronics—including remotes, modems, and motherboards—were iteratively expanded and classified using models such as SVM, Random Forest, Gradient Boosting, PLSDA, and Logistic Regression, achieving macro F1 scores nearing 1.0. The method revealed widespread plastic iron oxide, cerium-rich allanite, and hazardous compounds like benzanthracene. PCA plots and confusion matrices confirmed robust separability. This non-destructive and scalable approach advances hazard-aware sorting of heterogeneous e-waste and supports critical material recovery, offering a path toward intelligent, policy-aligned recycling infrastructure in a circular economy framework. |