AUTOMATED SURFACE INSPECTION FOR INDUSTRIAL APPLICATIONS

Abstract

Automated Surface Inspection (ASI), which applies computer vision algorithms for product quality control, has always been an important research field in manufacturing industry. The existing ASI methods largely automate the surface inspection processes that are traditionally performed by human experts. The primary aim of this Ph.D. project is to deal with the current challenges of ASI, and close the gap between academic ASI methods and industrial surface inspection. The proposed methods are divided into two parts. The first part focuses on local surface anomalies, where methods for texture classification, multiclass ASI, and weak and noisy labels are proposed. The second part focuses on global abnormalities, where a method using region-based graph is proposed for titanium alloy inspection. The experimental results on multiple public and industrial datasets show that the proposed methods efficiently solve various industrial ASI problems.