ADAMNet: Improving imbalanced defect classification with Anomaly-Driven Attention Maps

doi:10.1016/j.measurement.2026.120537

ADAMNet: Improving imbalanced defect classification with Anomaly-Driven Attention Maps

2026-02-05

0

PRE

AI

DOI：10.1016/j.measurement.2026.120537

Original

Original request for help

Save

Abstract

En 中文

• Proposed ADAMNet: a PatchCore-based framework for 3-class defect detection. • Introduced sampling using Wide ResNet-50 and adaptive pooling features. • Achieved 90%+ accuracy on imbalanced data with 15.2% gain over baseline. • Provided quantitative and qualitative results proving real-world effectiveness.

Keywords:

ADAMNet

defect detection

imbalanced data

anomaly-driven attention

PatchCore

ADAMNet: Improving imbalanced defect classification with Anomaly-Driven Attention Maps

Abstract

Journal

Organization