E7.2 - Cutout as Augmentation in Constrative Learning for Detecting Burn Marks in Plastic Granules
- Event
- SMSI 2023
2023-05-08 - 2023-05-11
Nürnberg - Band
- Lectures
- Chapter
- E7 - New Measurement and Sensor Approaches
- Author(s)
- M. Jin, M. Heizmann - Karlsruher Institut für Technologie, Karlsruhe (Germany)
- Pages
- 274 - 275
- DOI
- 10.5162/SMSI2023/E7.2
- ISBN
- 978-3-9819376-8-8
- Price
- free
Abstract
Burn marks in plastic granules are formed during the plastic injection process. The granules with burn marks are not acceptable for use in industrial application and should be filtered out in a sorting process. AI-based anomaly detection approaches are widely used in area of visual-based sorting due to the high accuracy and the low requirement of expert knowledge. In this contribution, we show that using cutout, a simple data augmentation strategy, can improve the accuracy of a contrastive learningbased anomaly detection method. In this work, synthetic image data are used due to the lack of real data.