Low-Abundance Contaminant Detection Using Hyperspectral Imaging

Abstract

Low-abundance contaminant detection is difficult because target materials may occupy only a small portion of a sample and produce spectral signatures that are weaker than natural biological variation. Hyperspectral imaging provides a high-dimensional sensing framework for identifying these subtle contamination patterns in meat products. This topic focuses on weak-signal detection using spectral unmixing, abundance estimation, matched filtering, and machine learning-based spectral interpretation. A benchmark dataset for meat contamination detection is especially important because it enables consistent testing of detection sensitivity, false alarm control, and generalization. The literature also highlights the importance of separating contaminant abundance from background tissue variation, which is closely related to remote sensing mixed-pixel analysis. Such methods support non-destructive, early-stage contamination monitoring for food safety systems.