Defect Classification Using Machine Learning Techniques for Flat Display Panels

Abstract

Defect classification for a flat display panel (FDP) is the crucial process that identifies and classifies defects automatically during the final step of its manufacturing process. It plays an important role since it prevents possible malfunction by inspecting defects timely and reduces time for identifying inferior products. In this paper, we propose the defect classification methods for FDP using various machine learning techniques and provide the comparison among them for practical use in production environment. First, we identify defects through Gaussian filter and threshold technique. Then, those defects are classified into different types based on geometric characteristics of them using four machine learning techniques that are widely used. The experimental results using training and test sets of FDP images show considerable effectiveness in classifying defect types. We also believe that the comparison result might be quite useful when engineers determine methods for defect classification during FDP manufacturing.