Hyperspectral Unmixing for Weak Signal Meat Contamination Detection

Abstract

Weak signal detection is a central challenge in hyperspectral food inspection because contaminant signals may be spatially small, spectrally subtle, and easily obscured by background variation in meat tissues. This topic focuses on hyperspectral unmixing methods for detecting low-abundance meat contamination through spectral decomposition, endmember extraction, and abundance estimation. A benchmark-oriented perspective is important because weak contamination detection requires datasets and evaluation protocols that can distinguish between algorithmic sensitivity and overfitting to specific laboratory conditions. Hyperspectral unmixing provides a natural framework for separating contaminant spectra from complex biological backgrounds, while machine learning and spectral-spatial modeling improve robustness under noise, illumination variation, and sample heterogeneity. The literature also connects food safety inspection with broader remote sensing and biomedical imaging methods, where weak signals and mixed pixels are common analytical challenges.