A Survey on Unsupervised Anomaly Detection Algorithms for Industrial Images

Yajie Cui,Zhaoxiang Liu,Shiguo Lian

DOI: https://doi.org/10.1109/ACCESS.2023.3282993

2023-06-13

Abstract:In line with the development of Industry 4.0, surface defect detection/anomaly detection becomes a topical subject in the industry field. Improving efficiency as well as saving labor costs has steadily become a matter of great concern in practice, where deep learning-based algorithms perform better than traditional vision inspection methods in recent years. While existing deep learning-based algorithms are biased towards supervised learning, which not only necessitates a huge amount of labeled data and human labor, but also brings about inefficiency and limitations. In contrast, recent research shows that unsupervised learning has great potential in tackling the above disadvantages for visual industrial anomaly detection. In this survey, we summarize current challenges and provide a thorough overview of recently proposed unsupervised algorithms for visual industrial anomaly detection covering five categories, whose innovation points and frameworks are described in detail. Meanwhile, publicly available datasets for industrial anomaly detection are introduced. By comparing different classes of methods, the advantages and disadvantages of anomaly detection algorithms are summarized. Based on the current research framework, we point out the core issue that remains to be resolved and provide further improvement directions. Meanwhile, based on the latest technological trends, we offer insights into future research directions. It is expected to assist both the research community and industry in developing a broader and cross-domain perspective.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

This paper mainly reviews and analyzes the problem of unsupervised anomaly detection in industrial images.

### Problems the Paper Attempts to Solve

1. **Current Algorithm and Technology Deficiencies**: Most existing deep learning-based anomaly detection methods rely on supervised learning, which requires a large amount of labeled data, consuming significant human resources and being inefficient. Additionally, due to the scarcity of defect samples, the application of these methods is limited.

2. **Proposing Unsupervised Learning Methods**: To address the above issues, the paper focuses on the application of unsupervised learning methods in the field of industrial image anomaly detection. Unsupervised learning does not require labeled data, which greatly reduces the cost of data collection and improves the practicality of the methods.

3. **Method Classification and Comparison**: The paper classifies and summarizes the unsupervised anomaly detection algorithms proposed in recent years, including five types of methods: reconstruction-based methods, normalizing flow-based methods, representation-based methods, data augmentation-based methods, and algorithm enhancement. It also describes their innovations and frameworks in detail.

4. **Introduction of Public Datasets**: To evaluate the effectiveness of different methods, the paper also introduces public datasets available for industrial anomaly detection.

5. **Challenges and Future Directions**: The paper points out the core issues in current research, such as the contradiction between the False Alarm Rate (FAR) and the Missed Alarm Rate (MAR), and proposes directions for further improvement. Additionally, based on the latest technological trends, the paper provides insights into future research directions.

Through this work, the paper aims to provide a broader cross-disciplinary perspective for the research community and industry, promoting the development of intelligent manufacturing systems.