Our FR5 inspection planning set contains one faceplate photographed at several small angles. The bright streak moves, and three reviewers labelled the images differently without examining the physical part. I do not think those labels are a sound basis for tuning detection. How would you rebuild this small reference set?
Get quality to examine the actual faceplate and locate any real damage first. Keep each image tied to that piece and its presentation. Then photograph known accepted and rejected parts under repeatable conditions. Otherwise you're teaching the system to reproduce a disagreement about reflections.
17分