FH-Net: Frequency-domain and hierarchical dual enhancement for few-shot surface defect segmentation
Lili Fan, Zhanglin Yang, Fan Guo, Ping Fang · 2026
Read FH-Net: Frequency-domain and hierarchical dual enhancement for few-shot surface defect segmentation by Lili Fan, Zhanglin Yang, Fan Guo, Ping Fang (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/pmc-13581219.txt.
Description
Fan, Yang, Guo, and Fang published in twenty twenty-six a method for finding factory surface defects from very few labeled examples. Labeling defect images is costly, so factories often have only a handful. Few-shot methods learn new defect types from a few support images. Past methods overfit those few images and miss multi-scale detail. Their network, FH-Net, attacks both problems at once. One module perturbs frequency patterns while preserving meaning, enriching the support set.
A second module merges features from several levels with attention-based weighting. That strengthens representation of defects at different sizes. Tests on two public defect data sets beat recent rival methods. Both one-shot and five-shot settings showed the gains. Removing each module lowered performance, confirming both contribute. The work targets real inspection lines where examples are scarce.
Catalog record
| CREATOR | Lili Fan, Zhanglin Yang, Fan Guo, Ping Fang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 79 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |