Hyperspectral Imaging for Identifying Foreign Objects on Pork Belly.

Area of Science:

• Food Science and Technology
• Spectroscopy
• Computer Vision

Background:

• Food safety and quality control in processing industries face challenges with contaminant detection.
• Traditional visual inspection methods often fail to identify subtle contaminants.

Purpose of the Study:

• To assess the feasibility of hyperspectral imaging (HSI) for detecting foreign objects on pork belly meat.
• To develop an automated system for real-time quality control in food processing.

Main Methods:

• Utilized a Specim FX17 hyperspectral camera for data acquisition in the 900-1700 nm range.
• Implemented a segmentation approach using a lightweight Vision Transformer with pre- and post-processing.
• Tested the system on images with and without contaminants, including under varying production-line temperatures.

Main Results:

• The method achieved high detection accuracy, retaining most true positives and eliminating false positives on clean images.
• The system demonstrated robustness against noise, temperature drift (10-55 °C), and spectral similarity.
• Zero false positives were observed on a challenging set of 95 images with fat-like spectra.

Conclusions:

• Hyperspectral imaging is a feasible technology for automated contaminant detection in pork processing.
• The developed Vision Transformer-based approach offers high accuracy and training efficiency.
• This HSI system supports real-time quality control, significantly improving food safety assurance.