M. Sc. Tom Sander

My research focus lies on anomaly detection for irregularly structured surfaces. This involves using color and hyperspectral images to detect outliers on surfaces. The challenge in this task is that the images of the surfaces represent non-repeating and irregularly structured surfaces, so any new surface is unknown and unpredictable. Furthermore, the anomalies should be detected with as few as possible or even no labeled data points at all, so that the learning methods are limited to the following four:

• unsupervised
• weakly-supervised
• self-supervised
• semi-supervised

In the current application example, defects on natural wooden floors and furniture surfaces are to be detected in this way. This scenario is intended to enable the system to improve its decision-making boundary by using individual inputs from a domain expert.

• Sander, T., Lange, S., Hilleringmann, U., Geneiß, V., Hedayat, C., & Kuhn, H., 2022. Detection of Defects on Irregularly Structured Surfaces using Supervised and Semi-Supervised Learning Methods. In 2022 Smart Systems Integration (SSI) (pp. 1-6). IEEE. 10.1109/ssi56489.2022.9901433
• Sander, T., Lange, S., Hilleringmann, U., Geneiß, V., Hedayat, C., Kuhn, H., & Gockel, F. B., 2021. Detection of Defects on Irregular Structured Surfaces by Image Processing Methods for Feature Extraction. In 22nd IEEE International Conference on Industrial Technology (ICIT). Vol. 1. IEEE. 10.1109/icit46573.2021.9453646