A non-destructive, autoencoder-based approach to detecting defects and contamination in reusable food packaging

Area of Science:

• Food safety
• Environmental sustainability
• Computer vision

Background:

• The food service industry seeks sustainable practices, with reusable foodware adoption increasing.
• Manual inspection of reusable foodware for defects and contaminants is labor-intensive and error-prone.
• Automated detection is crucial for safety, scalability, and cost-efficiency in reusable foodware systems.

Purpose of the Study:

• To develop an automated method for detecting defects and contaminants on reusable food containers.
• To improve the safety and efficiency of reusable foodware systems.
• To enable rapid deployment of detection systems with limited data.

Main Methods:

• Utilizing data augmentation strategies.
• Applying knowledge transfer from diverse food container samples.
• Training a network to identify normal patterns and detect deviations indicating defects or contaminants.

Main Results:

• The approach effectively detects both contamination and cracks on food containers.
• The system can be rapidly deployed even with a limited number of collected samples.
• The method leverages few images of clean samples to teach the network normal patterns.

Conclusions:

• Automated detection systems are vital for ensuring consumer safety in reusable foodware systems.
• The proposed method offers a scalable and cost-efficient solution for defect and contaminant detection.
• This research contributes to advancing sustainable practices in the food service industry through technology.