Case Study: Reading the Best Before Date on Yogurt Cups Using Deep Learning

Industry Challenge

Reading expiration dates on yogurt containers is a challenge for traditional vision systems due to several factors:

• Distorted Characters: The expiration date is printed on a curved surface, leading to distortions that make it difficult for conventional systems to capture and interpret the text.
• Varying Font Sizes and Contrast: Different containers have inconsistent font sizes and sometimes low contrast between the ink and the background, especially in inkjet printing.
• Lighting Conditions: Varying light levels, reflections, and shadows can obscure the printed information, reducing readability.

Application of Deep Learning

senswork used Deep Learning to provide a solution to these challenges by allowing for adaptive learning from various examples of yogurt containers, ensuring reliable detection in less-than-ideal conditions. Here are specific tasks we addressed:

1. Accurate Detection
   Deep Learning models are trained on large datasets of images showing expiration dates on yogurt cups, enabling them to accurately detect and localize the text, even on curved surfaces with distorted fonts.

2. Legibility and Completeness
   By using deep neural networks, the system checks the clarity and completeness of each character, ensuring that the entire date is visible and interpretable.

   ---

   SUBSCRIBE TO A3 NEWSLETTERS

   The A3 newsletters are your gateway to the latest industry updates, expert insights, and exclusive content in robotics, vision & imaging, motion control & motors, and industrial artificial intelligence technologies.

   Subscribe to A3 Newsletters

   ---

    

3. Fast and Adaptable Reading
   The system is optimized for speed, enabling real-time detection on production lines where multiple yogurt cups are processed rapidly. It adapts to different backgrounds without requiring manual tuning, unlike traditional machine vision systems.

4. Handling Low Contrast and Distortions
   senswork's Deep Learning model excels at recognizing characters even with low contrast or varying lighting conditions. This is particularly useful for inkjet-printed dates, where contrast may be low and font distortions are common.

Outcomes 

• Increased Accuracy: Deep Learning significantly improved accuracy in detecting and reading expiration dates on yogurt cups under challenging conditions.
• Reduced Errors: The system ensures that expiration dates are reliably read, reducing the chances of products with missing or incorrect date readings entering the market.
• Scalability: This solution can also be adapted to other product types and packaging formats that use inkjet printing for expiration dates.

Deep Learning highlights how senswork overcomes the limitations of traditional vision systems in reliably reading best-before dates on curved, low-contrast, and distorted surfaces, ensuring compliance and efficiency in the production process.