EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Calibrated Feature Fusion: Enhancing Few-Shot Industrial Anomaly Detection via Cross-Stage Representation Alignment

Shuangjun Zheng iD, Songtao Zhang, Zhihuan Huang, Kuoteng Sun, Yuzhong Gong iD, Jiayan Wen iD, Eryun Liu iD

Sensors · 31.03.2026 · 26 (7) · 2164

DOI: 10.3390/s26072164 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Sensors, online erschienen am 31.03.2026. Verfasst von Shuangjun Zheng und 6 weiteren Autor:innen. Beteiligte Einrichtung: College of Information Science & Electronic Engineering, Zhejiang University, Hangzhou 310027, China. Die Arbeit nennt 47 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Zhejiang Provincial Natural Science Foundation of China, the Zhejiang Provincial Natural Science Foundation, the “Leading Goose” R&D Program of Zhejiang.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Few-shot industrial anomaly detection technology has received more and more attention because it does not require a large number of abnormal samples to train. Recent few-shot industrial anomaly detection methods commonly fuse multi-stage features from frozen vision transformers for anomaly scoring. However, we find that such direct fusion suffers from cross-stage representation misalignment—shallow and deep features differ significantly in scale and semantic granularity, leading to inconsistent anomaly maps and degraded localization. To address this problem, we propose Calibrated Feature Fusion (CFF), a lightweight adapter that enhances feature fusion via cross-stage representation alignment. The CFF module can be integrated into existing state-of-the-art frameworks and operates effectively in few-shot settings. Experiments on MVTec AD and VisA show that CFF consistently improves the state-of-the-art method across 1/2/4-shot settings, achieving gains of up to +1.6% AUROC and +4.1% AP in pixel-level segmentation. Notably, CFF enhances both precision and recall in four-shot scenarios. Ablation studies confirm that cross-stage alignment is key to stable multi-stage fusion.

Schlagwörter

Stichwörter der Autor:innen

cross-domain adaptationcross-stage alignmentfeature fusionfew-shot learningindustrial anomaly detectionvision transformers

Autor:innen und Institutionen

  • College of Information Science & Electronic Engineering, Zhejiang University, Hangzhou 310027, China.
  • Liuzhou Bureau, Extra High Voltage Transmission Company, China Southern Power Grid Co., Ltd., Liuzhou 545026, China.
  • College of Control Science and Engineering, Zhejiang University, Hangzhou 310027, China.
  • School of Automation (School of Artificial Intelligence), Guangxi University of Science and Technology, Liuzhou 545006, China.

ORCID-Profile: Shuangjun Zheng, Yuzhong Gong, Jiayan Wen, Eryun Liu

Förderung

  • Zhejiang Provincial Natural Science Foundation of China
  • the Zhejiang Provincial Natural Science Foundation
  • the “Leading Goose” R&D Program of Zhejiang

Zitieren

APA

Zheng, S., Zhang, S., Huang, Z., Sun, K., Gong, Y., Wen, J., & Liu, E. (2026). Calibrated Feature Fusion: Enhancing Few-Shot Industrial Anomaly Detection via Cross-Stage Representation Alignment. Sensors, 26(7), 2164. https://doi.org/10.3390/s26072164

Vancouver

Zheng S, Zhang S, Huang Z, Sun K, Gong Y, Wen J, et al. Calibrated Feature Fusion: Enhancing Few-Shot Industrial Anomaly Detection via Cross-Stage Representation Alignment. Sensors. 2026;26(7):2164. doi:10.3390/s26072164

BibTeX
@article{Zheng2026,
  author = {Zheng, Shuangjun and Zhang, Songtao and Huang, Zhihuan and Sun, Kuoteng and Gong, Yuzhong and Wen, Jiayan and Liu, Eryun},
  title = {Calibrated Feature Fusion: Enhancing Few-Shot Industrial Anomaly Detection via Cross-Stage Representation Alignment},
  journal = {Sensors},
  year = {2026},
  volume = {26},
  number = {7},
  pages = {2164},
  doi = {10.3390/s26072164}
}
Kostenlos starten →