Paradigm Evolution of Industrial Surface Defect Detection: The Underlying Logic and Fundamental Challenges from Supervised Classification to Unsupervised Anomaly Localization

DOI

10.2991/978-94-6239-648-7_75How to use a DOI?

Keywords

Industrial appearance defect detection; Fully supervised learning; Unsupervised anomaly localization; Open-vocabulary detection

Abstract

Industrial appearance defect detection is a key aspect of quality control in smart manufacturing. The core challenge lies in how to effectively apply models that perform well in laboratory environments to complex, dynamic, and unpredictable real-world industrial scenarios. This challenge is specifically manifested in the sharp contradiction between the infinite possibilities of defect types in industrial production lines and the extreme scarcity of labeled data. This paper systematically reviews the paradigm evolution in this field, from closed-set fully supervised classification to open-set anomaly detection, and further to open-vocabulary semantic understanding. First, an in-depth analysis was conducted on the inherent limitations of fully supervised models based on U-Net and ResNet regarding label dependency and distribution shift issues. Secondly, it explores how unsupervised paradigms (such as PatchCore and PaDiM) achieve unknown defect detection by learning only normal samples, as well as their shortcomings in distinguishing defect types. Finally, it describes how the open-vocabulary paradigm based on vision-language large models (such as CLIP) provides new approaches for zero-shot defect classification through semantic guidance. By comparing the performance and underlying logic of various paradigms on benchmark datasets such as MVTec AD and VisA, this paper reveals the development trends in the field of industrial visual inspection and provides theoretical guidance and practical directions for building the next generation of highly adaptable and robust industrial detection systems.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Zihan Zhang
PY  - 2026
DA  - 2026/04/24
TI  - Paradigm Evolution of Industrial Surface Defect Detection: The Underlying Logic and Fundamental Challenges from Supervised Classification to Unsupervised Anomaly Localization
BT  - Proceedings of the International Workshop on Advances in Deep Learning for Image Analysis and Computer Vision (IWADIC 2025)
PB  - Atlantis Press
SP  - 693
EP  - 701
SN  - 2352-538X
UR  - https://doi.org/10.2991/978-94-6239-648-7_75
DO  - 10.2991/978-94-6239-648-7_75
ID  - Zhang2026
ER  -