Presentation Information

5:15 PM - 5:30 PM JST(8:15 AM - 8:30 AM UTC)

[10p-A33-13]Pinhole detection in opaque containers using polarized light and machine learning

〇KENZO YAMAGUCHI^1,2,3, MASAMI SHISHIBORI², RYO HADA³, TOSHIYUKI KASHIWAGI³, HITOSHI OGAWA³ (1.NIT, Anan College, 2.Tokushima Univ., 3.Tokushima Pref. ITC)

Keywords:

Polarized Transmitted Light,Pinhole Detection,Machine Learning

As food containers become lighter and thinner, reliable inspection of pinhole defects generated during the forming process is increasingly important. In this study, pinhole detection in opaque food containers containing white talc was investigated using polarized transmitted light and machine learning. Hole-related features were extracted through inverted-difference processing of parallel- and crossed-Nicol images, background correction using a superimposed image generated from 100 hole-free containers, and tone transformation. A VGG16-based classifier achieved a recall of 1.000 and a precision of 0.976. Furthermore, automated inspection was successfully demonstrated by integrating the detection system with a SCARA robot for container handling.

Back to Session information