The challenge
What needed solving
Photographs varied in angle, orientation, lighting and packet design. The code fields had to be located before their characters could be read.
The solution
How Qualitas solved it
An object-detection model located text regions and character-recognition software read them. Volunteers uploaded images and checked the extracted fields.
How the system works
YOLO located the date and serial regions; PaddleOCR read the cropped text. Separating field location from reading supported photographs captured without a fixed factory fixture.
Results and practical value
Date extraction reached 99.14% accuracy and serial-code recognition reached 95.66% after three months of development.
Share representative photographs, fields to extract, expected formats and the verification workflow.
Request project details
