Hyperspectral imaging refers to the combination of spectroscopy and digital image processing. This relatively new imaging technology makes it possible to measure or infer and visualize the chemical properties of an object spatially resolved. For this purpose, methods of machine learning and multivariate data analysis must be used to model the relationship between the spectroscopic measurement data and the chemical properties. In particular, current approaches to deep learning (eg representation learning, transfer learning and autoencoders) and anomaly detection are investigated and applied. Anomaly detection is an increasingly relevant area of research especially in the context of "food fraud detection".
Developement and Application of machine learning methods for spectroscopic image and data analysis
|Motion-based material characterization in sensor-based sorting||Maier, G.; Pfaff, F.; Becker, F.; Pieper, C.; Gruna, R.; Noack, B.; Kruggel-Emden, H.; Längle, T.; Hanebeck, U.; Wirtz, S.; Scherer, V.; Beyerer, J.||tm - Technisches Messen 85 no. 3, pp. 202-210, De Gruyter, Oldenbourg, 2018.|
|Improving Material Characterization in Sensor-Based Sorting by Utilizing Motion Information (to appear)||Maier, G.; Pfaff, F.; Becker, F.; Pieper, C.; Gruna, R.; Noack, B.; Kruggel-Emden, H.; Längle, T.; Hanebeck, U.; Wirtz, S.; Scherer, V.; Beyerer, J.||KIT Scientific Publishing, K. (ed.), Proceedings of the 3rd Conference on Optical Characterization of Materials (OCM 2017), 2017.|