Visual Anomaly Detection and Applications in Product Quality Assessment

Currently, most small or medium manufacturers inspect products or parts’ visual quality manually. The manual visual quality inspection is error prone, subjective, and labor intensive. In recent years, image-based automated
inspection has seen exciting uptakes in many modern factories. One critical challenge that limits their vast potentials lies in its ability to detect anomalous situations and events that usually involve product defects in
manufacturing processes – the central theme of our proposed project. In particular, we list the following key objectives: 1) learning anomaly detection with positive instances; 2) unsupervised learning anomaly detection;
3) a practical pipeline of visual anomaly detection, with a focus on in metal bottle manufacture. Our research program aims to address these fundamental and practical challenges, as well as to make practical impact in
real-world applications.

Faculty Supervisor:

Li Cheng

Student:

Partner:

Zerobox

Discipline:

Computer science

Sector:

Professional, scientific and technical services

University:

University of Alberta

Program: