C1 Learning Based Analysis of Multispektral THz-Data

Projektleaders

Prof. Dr. Volker Blanz, Prof. Dr.-Ing. Andreas Kolb

Initial Situation

With the development of imaging and spectroscopic measuring technology in the THz range, the question of data analysis becomes more and more important. In contrast to visible light, other physical effects become important, which considerably complicate material recognition in the spectra, as well as object recognition in image data. Such is the case e.g. with overlapping of signals from different materials, .

Aims and Work Program

The aim of the project is to synergetically use the existing expertise in the field of THz measuring technology, including a broad palette of measuring methods, and of object recognition and -detection. Thus, current methods of mechanical vision and mechanical learning of data can be applied to the respective latest available THz measuring technologies, in order to gather information about the acquired objects and materials. Concrete aims are the system independent recognition of materials, recognition of objects in current image data, and spectral analysis through an Analysis-by-Synthesis approach.  In contrast to sub-project D2, this approach processes multispectral THz-data, which are captured by means of THz-detection methods.
Within the framework of the Research Training Group a method of material recognition, based on mechanical learning, has been developed. This approach is, like other methods of automatic material recognition, tied to specific measuring systems. These detection processes are not based on system independent material data, but rather model the spectral system characteristic of the measurement assembly and the sample geometric dependencies implicitly. Based upon these results, new, mostly system independent methods are to be developed, which use the developed proceedures for extraction of material parameters, in order to separate the properties of the measuring system from the characteristic of the measured materials.

Absorption of TNT samples of different manufacturers.	Top view of the imaging geometry of a THz-Scanner System