Unsupervised Foreign Object Detection Based on Dual-Energy Absorptiometry in the Food Industry.

Area of Science:

• Food Science and Technology
• Imaging and Sensing Technologies
• Quality Control and Assurance

Background:

• X-ray imaging is crucial for non-destructive inspection of agricultural food products.
• Autonomous, in-line detection of foreign objects (e.g., bone, plastic, metal, infestations) is vital for food safety.
• Current methods face challenges with noise and contrast, impacting detection accuracy.

Purpose of the Study:

• To develop an unsupervised foreign object detection methodology for food products using dual-energy X-ray absorptiometry (DEXA).
• To introduce a novel thickness correction model for DEXA data pre-processing to enhance foreign object contrast and segmentation robustness.
• To validate the proposed methodology on a real-world dataset of meat products.

Main Methods:

• Utilized dual-energy X-ray absorptiometry (DEXA) for X-ray imaging.
• Developed and applied a novel thickness correction model as a pre-processing step for DEXA data.
• Implemented an unsupervised foreign object detection algorithm.
• Tested the methodology on 488 meat product samples from a conveyor belt.

Main Results:

• The thickness correction model effectively homogenized food product regions and enhanced foreign object contrast.
• Samples without foreign objects were correctly identified in 97% of cases.
• The overall accuracy of foreign object detection reached 95%.

Conclusions:

• The proposed DEXA-based methodology with thickness correction is effective for unsupervised foreign object detection in food products.
• The technique offers a robust and accurate solution for in-line quality control in the food industry.
• This approach significantly improves the reliability of detecting contaminants in meat products.