Foreign objects in food pose significant risks to consumer safety. The discovery of physical contaminants often results in costly product recalls, severe financial losses, and long-term damage to a brand’s reputation.
To ensure the highest levels of quality assurance and food safety, manufacturers must implement robust inspection systems throughout the production process. Traditional visual inspections and standard RGB cameras frequently fail when contaminants match the color or texture of the food itself. Our integrated Near-Infrared (NIR) hyperspectral imaging solutions offer a definitive, automated answer to this critical industry pain point.
To demonstrate the capability of our systems, an inspection study was conducted across three distinct food categories that carry high processing risks: chicken fillets, veggie patties, and goat cheese. By placing common factory-floor contaminants on each product, a classification model was developed to evaluate real-time automated detection.
Testing Scenarios: Three Challenging Food Matrices
Poultry Inspection (Chicken Fillet)
First, the system evaluated poultry processing, a high-value sector with strict safety compliance requirements. Common industrial contaminants—including wood fragments, metal pieces, and two types of widely used plastics (Polyethylene – PE and Polystyrene – PS)—were introduced onto the product stream.
While manual inspection can easily miss clear plastics or small wood splinters on raw meat, the hyperspectral system maps the unique molecular chemistry of each object instantly.
Figure 1: Factory-floor automated core inspection layout during spectral scanning.
Plant-Based Alternatives (Veggie Patties)
Next, the inspection setup examined plant-based veggie patties using the same set of foreign materials (wood, metal, PE, and PS plastics). The complex, textured, and colorful surface of vegetable-based matrices makes traditional vision systems highly unreliable, as food ingredients can easily be mistaken for contaminants.
Figure 2: Veggie patties with introduced contaminants: Visual photo (left) vs. hyperspectral identification mapping (right).
High-Contrast Challenges (Goat Cheese Packaging)
Finally, the system was tested against one of the most difficult challenges in food packaging: detecting thin, transparent plastic wrapping on goat cheese. Because the thin white film looks almost identical to the cheese surface under normal lighting, it remains completely invisible to the human eye and traditional RGB cameras.
Figure 3: Goat cheese sample with a transparent plastic packaging contaminant: Standard photo (left) vs. clear hyperspectral false-color detection (right).
Spectral Analysis: Distinguishing Food from Foreign Material
Our integrated processing software normalizes the raw reflectance data against precise white and dark references to establish clean chemical signatures. By analyzing the mean spectrum of both the food matrices and the introduced foreign objects, the data demonstrates that every material possesses a unique, unmistakable spectral signature.
What We Solve in Practice:
- Unmistakable Chemical Profiling: The spectral signatures of meat, poultry, and vegetable proteins differ fundamentally from industrial polymers and metals. Even when colors are identical, the system separates them based on chemical composition.
- Overcoming Transparent Materials: In complex cases like goat cheese packaging, thin films can be slightly transparent, causing the spectral signatures of the food and the plastic to blend. Standard systems fail here. However, our advanced classification models isolate these overlapping signatures, highlighting the contaminant profile even when it is physically embedded or obscured.
Figure 4: Spectral signature comparison separating poultry tissue from PS, PE, wood, and metal.
Figure 5: Spectral signature comparison isolating vegetable-based matrices from factory contaminants.
Figure 6: Spectral differentiation showing the isolated chemical contrast between goat cheese and thin packaging film.
Machine Learning & Automated Classification
To turn this raw spectral data into instant factory actions, a customized Partial Least Squares Discriminant Analysis (PLS-DA) classification model was built for each application.
For the chicken fillets and veggie patties, the automated algorithm was trained to sort five distinct classes simultaneously (PE, PS, wood, metal, and the food product). For the goat cheese line, the model focused purely on binary sorting (isolating the packaging film from the cheese). The processing system automatically detects and masks the conveyor background in black, leaving a clean, color-coded map for the sorting rejection mechanism.
Figure 7 Automated classification results for poultry inspection showing isolated contaminant zones.
Figure 8: Veggie patty sorting model accurately identifying multiple material types simultaneously.
Figure 9: Goat cheese packaging detection model highlighting the thin, otherwise invisible plastic film in red.
Conclusion
Integrating advanced NIR hyperspectral imaging into food production lines completely transforms quality control from reactive damage control into proactive, 100% automated safety assurance. The findings of this study prove that:
- The spectral signatures of diverse food products and industrial contaminants are highly distinguishable.
- Our integrated classification models reliably automate the real-time sorting and rejection of foreign objects based on automated infrared data.
- Our hyperspectral systems successfully detect and reject critical contaminants that are entirely invisible to the human eye and standard factory vision hardware.
