Robust Spectral Feature Extraction for Contamination Detection

Abstract

Robust spectral feature extraction is essential for contamination detection when target signals are weak, noisy, or partially mixed with background material. In meat safety inspection, hyperspectral images contain complex spectral variation caused by tissue composition, moisture, fat distribution, illumination, and acquisition geometry. This topic focuses on methods that extract reliable spectral features from weak contamination signals using unmixing, preprocessing, dimensionality reduction, and machine learning. Benchmark datasets are valuable because they support systematic comparison of feature extraction methods under controlled weak-signal conditions. Robust features should preserve subtle contamination signatures while suppressing irrelevant variation. The literature connects food inspection with broader hyperspectral image processing methods, including spectral mixture analysis, sparse representation, and spectral-spatial modeling.