Computer vision quality inspection

The food technology company inspected packaging units manually on the line — slow and error-prone. The goal was automated visual inspection without stopping the line.

RALUS installed three industrial cameras and trained a CNN model on 12,000 labelled images. Defective units are detected with 97% accuracy and rejected automatically — without stopping production.

After go-live, the defect rate in the end product fell below 0.3%. Manual incoming inspection was reduced by 91%.