Beer-Lambert Guided Representation Learning for Unsupervised Anomaly Detection in Sub-THz Food Inspection Images
2026-06-15 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors address the challenge of spotting unusual stuff in food using sub-THz imaging, which can see inside food by how much different materials block the signal. They note that current methods use image features designed for regular photos and may miss important details in these special images. To fix this, they created a new way to teach the computer by helping it understand how the sub-THz signal weakens inside the food, based on the Beer-Lambert law. They also tested their method on a dataset with different food types and showed it detects problems better than older methods.
Sub-THz transmission imagingAnomaly detectionBeer-Lambert lawRepresentation learningUnsupervised learningAttenuation decompositionFood inspectionOne-class settingGeneralizationInline-Food-Inspection-THz dataset
Authors
Gyutae Hwang, Sang Jun Lee
Abstract
Food manufacturing requires reliable inspection systems to detect foreign material contamination and maintain product safety. Sub-THz transmission imaging provides material-dependent attenuation characteristics that are useful for detecting low-density contaminants in food products. However, existing unsupervised anomaly detection methods mainly rely on RGB-pretrained visual representations, which may not adequately capture the transmission behavior of Sub-THz images. This paper proposes a Beer-Lambert guided representation learning framework for unsupervised anomaly detection in Sub-THz food inspection images. The proposed method introduces an attenuation decomposition module as an auxiliary regularization module that constrains student representations through attenuation reconstruction during training. In addition to the conventional one-class setting, we introduce a Leave-One-Food-Out protocol to evaluate generalization capability under unseen food categories. Experimental results on the Inline-Food-Inspection-THz dataset show that the proposed method improves overall anomaly detection performance over the baseline method.