A prototype-aligned multi-scale attention network for few-shot industrial surface defect detection
Rubing Huang, Lianyi Zhao, Xiaowei Li · 2026
Read A prototype-aligned multi-scale attention network for few-shot industrial surface defect detection by Rubing Huang, Lianyi Zhao, Xiaowei Li (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/PMC13618933.txt.
Description
Rubing Huang and two colleagues design a neural network that spots industrial surface defects from a handful of labeled examples, a chronic problem. Defect photos are scarce, defect sizes vary wildly, and textured metal backgrounds trigger false alarms. Their Prototype-Aligned Multi-Scale Attention Network fuses features at several scales. It applies attention over space and channels, and classifies regions by matching them against class prototypes inside a two-stage detector.
On the NEU-DET steel defect benchmark with five examples per class, the network reached mean average precision above ninety-two percent at the overlap threshold. It also showed strong small-object accuracy and a false response rate near four percent. It beat a Faster R-CNN baseline on every headline measure. Cross-domain testing on a second dataset, GC10-DET, held precision in the mid-eighties. The authors support region-level prototype alignment for limited-label metal inspection.
Catalog record
| CREATOR | Rubing Huang, Lianyi Zhao, Xiaowei Li |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 57 KB |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |