F-005
Glass Surface Defect Grading using Machine Learning Methods

◎Monikka Roslianna Busto・Takashi Obi（東工大）・Hiroyuki Suzuki（群馬大）・Joong Sun Lee・Pei Jiang（東工大）

Current studies on surface defect analysis involve detection of defects during manufacturing with little assessment of damage severity. In this study, we extend the defect analysis to grading - determining overall severity of damaged glass surfaces. We propose a framework for the computer vision tasks to perform grading and explore improvements to machine learning methods for defect detection. Furthermore, we aggregate information from detected defects and train a supervised machine learning classifier to assess severity. Experimental results demonstrate feasibility of defect grading using a minimal amount of labeled data.