Composite Score for Anomaly Detection in Imbalanced Real-World Industrial Dataset

TL;DR

This paper proposes a novel composite scoring method for anomaly detection in imbalanced industrial datasets, combining generative reconstruction and multi-level metrics to improve accuracy while satisfying zero-false-negative constraints.

Contribution

It introduces a three-step approach using VQGAN reconstruction, multi-level metric extraction, and classification to enhance anomaly detection in complex industrial images.

Findings

Achieves 95.69% accuracy on MVTec-AD dataset.

Attains 87.93% accuracy on PCBA dataset under ZFN constraint.

Demonstrates effectiveness of composite scoring in imbalanced industrial scenarios.

Abstract

In recent years, the industrial sector has evolved towards its fourth revolution. The quality control domain is particularly interested in advanced machine learning for computer vision anomaly detection. Nevertheless, several challenges have to be faced, including imbalanced datasets, the image complexity, and the zero-false-negative (ZFN) constraint to guarantee the high-quality requirement. This paper illustrates a use case for an industrial partner, where Printed Circuit Board Assembly (PCBA) images are first reconstructed with a Vector Quantized Generative Adversarial Network (VQGAN) trained on normal products. Then, several multi-level metrics are extracted on a few normal and abnormal images, highlighting anomalies through reconstruction differences. Finally, a classifer is trained to build a composite anomaly score thanks to the metrics extracted. This three-step approach is performed on the public MVTec-AD datasets and on the partner PCBA dataset, where it achieves a regular accuracy of 95.69% and 87.93% under the ZFN constraint.