Supervised Anomaly Detection for Complex Industrial Images

Aimira Baitieva,David Hurych,Victor Besnier,Olivier Bernard

2024-05-11

Abstract:Automating visual inspection in industrial production lines is essential for increasing product quality across various industries. Anomaly detection (AD) methods serve as robust tools for this purpose. However, existing public datasets primarily consist of images without anomalies, limiting the practical application of AD methods in production settings. To address this challenge, we present (1) the Valeo Anomaly Dataset (VAD), a novel real-world industrial dataset comprising 5000 images, including 2000 instances of challenging real defects across more than 20 subclasses. Acknowledging that traditional AD methods struggle with this dataset, we introduce (2) Segmentation-based Anomaly Detector (SegAD). First, SegAD leverages anomaly maps as well as segmentation maps to compute local statistics. Next, SegAD uses these statistics and an optional supervised classifier score as input features for a Boosted Random Forest (BRF) classifier, yielding the final anomaly score. Our SegAD achieves state-of-the-art performance on both VAD (+2.1% AUROC) and the VisA dataset (+0.4% AUROC). The code and the models are publicly available.

Computer Vision and Pattern Recognition,Machine Learning

What problem does this paper attempt to address?

This paper attempts to solve the problem of automated visual inspection of complex images on industrial production lines. Specifically, existing anomaly detection methods perform poorly when dealing with actual industrial images containing defects, mainly because existing public datasets are mainly composed of defect - free images, which limits the application effectiveness of these methods in actual production environments. To address this challenge, the paper makes two main contributions:

1. **Valeo Anomaly Dataset (VAD)**: This is a new real - world industrial dataset containing 5,000 images, of which 2,000 contain complex real - world defects in more than 20 sub - classes. This dataset aims to弥补 the deficiencies of existing datasets and provide more defect samples for training more effective anomaly detection models.

2. **Segmentation - based Anomaly Detector (SegAD)**: This is a new supervised anomaly detection method. SegAD calculates local statistical features by using anomaly maps and segmentation maps, and uses these statistical features and an optional supervised classifier score as input features, and finally generates the final anomaly score through a Boosted Random Forest (BRF) classifier. Experimental results show that SegAD has achieved state - of - the - art performance on both the VAD and VisA datasets (increasing the AUROC by 2.1% and 0.4% respectively).

The main objective of the paper is to improve the anomaly detection effect of complex images on industrial production lines by providing a new dataset and a new detection method, thereby improving product quality and production efficiency.