Related Experiment Video
Updated: Jul 10, 2026
05:47
Simulation of a Scaled Assembly Process with Collaboration of a Robotic Arm and Monitoring through a Vision System for Quality Control
Published on: August 29, 2025
Vision expert guided inspection for industrial anomaly detection
Xiangyu Zhu1,2, Wenhua Cui2,3, Ye Tao2,3
1School of Electronic and Information Engineering, University of Science and Technology Liaoning, Anshan, China.
Plos One
|July 8, 2026
Summary
This study introduces a novel visual expert-guided method for industrial anomaly detection (IAD), enhancing defect identification. The approach excels at finding subtle defects, improving product quality and equipment safety in intelligent manufacturing.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Manufacturing Engineering
Background:
- Industrial anomaly detection (IAD) is crucial for quality control and safety in intelligent manufacturing.
- Identifying subtle and generalized defects remains a challenge due to vague feature representation.
- Existing methods struggle with precise localization and comprehensive anomaly information extraction.
Purpose of the Study:
- To propose a visual expert-guided multi-scale anomaly detection method for enhanced defect identification.
- To improve the detection of subtle and generalized anomalies in industrial settings.
- To enhance the robustness and generalization capabilities of anomaly detection models.
Main Methods:
- Leveraging super-resolution techniques to enhance spatial resolution and recover fine-grained details for discriminative defect representation.
- Employing a graph attention network-based multi-scale fusion module to aggregate suspicious regions across scales, modeling inter-scale dependencies.
- Dynamically weighting and localizing features to preserve both micro irregularities and macro structural deviations for comprehensive anomaly information.
Main Results:
- The proposed method consistently outperforms existing approaches in both image-level and pixel-level anomaly detection across public datasets.
- Achieved high pixel-level accuracy (98.6% and 98.1% in 4-shot settings, 94.6% in zero-shot settings) on major benchmarks.
- Demonstrated strong capability in detecting subtle defects on fine-grained textures and enhanced robustness in cross-domain transfer scenarios.
Conclusions:
- The visual expert-guided multi-scale anomaly detection method significantly advances industrial defect detection capabilities.
- The integration of super-resolution and graph attention networks provides a powerful framework for identifying subtle and complex anomalies.
- The method offers improved accuracy, robustness, and generalization, contributing to safer and higher-quality intelligent manufacturing.