Unsupervised and Semi-Supervised Learning in Automatic Industrial Image Inspection

Weitao Tang, Purdue University

Abstract

It has been widely studied in industry production environment to apply computer version on X-ray images for automatic visual inspection. Traditional methods embrace image processing techniques and require custom design for each product. Although the accuracy of this approach varies, it often fall short to meet the expectations in the production environment. Recently, deep learning algorithms have significantly promoted the capability of computer vision in various tasks and provided new prospects for the automatic inspection system. Numerous studies applied supervised deep learning to inspect industrial images and reported promising results. However, the methods used in these studies are often supervised, which requires heavy manual annotation. It is therefore not realistic in many manufacturing scenarios because products are constantly updated. Data collection, annotation and algorithm training can only be performed after the completion of the manufacturing process, causing a significant delay in training the models and establishing the inspection system. This research was aimed to tackle the problem using unsupervised and semi-supervised methods so that these computer vision-based machine learning approaches can be rapidly deployed in real-life scenarios. More specifically, this dissertation proposed an unsupervised approach and a semi-supervised deep learning method to identify defective products from industrial inspection images. The proposed methods were evaluated on several open source inspection datasets and a dataset of X-Ray images obtained from a die casting plant. The results demonstrated that the proposed approach achieved better results than other state-of-the-art techniques on several occasions.

Degree

Ph.D.

Advisors

Yang, Purdue University.

Subject Area

Artificial intelligence|Computer science|Design|Industrial engineering|Materials science

COinS