Defect detection using weakly supervised learning

TL;DR

This paper tackles the challenge of labeled-data scarcity in defect detection by evaluating a positive-unlabeled (PU) weakly supervised framework. It leverages a frozen VGG-16 feature extractor to produce high-level representations, uses an Isolation Forest to rank anomalies and form a counter-example set, and trains a binary classifier on the positive set plus counter-examples. Across the Ball Screw Defect for Classification (BSD) dataset, the weakly supervised approach achieves substantial performance gains with limited labels, approaching the fully supervised model as the fraction of labeled data increases (roughly 80% accuracy at 5% labels to ~93% at 30%). The work demonstrates data-efficient defect detection with potential for integration with domain knowledge, transfer, or active learning to further enhance industrial inspection systems.

Abstract

In many real-world scenarios, obtaining large amounts of labeled data can be a daunting task. Weakly supervised learning techniques have gained significant attention in recent years as an alternative to traditional supervised learning, as they enable training models using only a limited amount of labeled data. In this paper, the performance of a weakly supervised classifier to its fully supervised counterpart is compared on the task of defect detection. Experiments are conducted on a dataset of images containing defects, and evaluate the two classifiers based on their accuracy, precision, and recall. Our results show that the weakly supervised classifier achieves comparable performance to the supervised classifier, while requiring significantly less labeled data.

Figures (2)

• Figure 1: Overview of the proposed method
• Figure 2: Random samples from the BSD dataset. Non-defective samples might exhibit lubrication signs raising a false alarm in defect detection systems.