Fraunhofer-Gesellschaft

Publica

Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Project network HuMin/MD - advanced data analysis methods for metal detectors

 
: Eigenbrod, H.

Nitsch, J.:
EUROEM 2004. Book of Abstracts : Euro Electromagnetics, 12 - 16 July 2004, Magdeburg, Germany
Magdeburg: Univ. Magdeburg, 2004
ISBN: 3-929757-73-7
Euro Electromagnetics (EUROEM) <2004, Magdeburg>
Englisch
Konferenzbeitrag
Fraunhofer IPA ()
Signalanalyse; measurement; Bildverarbeitung; Signalverarbeitung

Abstract
The BMBF-funded German research project on Humanitarian Demining (HuMin/MD) aims to reduce the number of false alarms produced by metal detectors. To achieve this, the project network will primarily develop the potential of secondary mathematical methods for analyzing data obtained from commercial off-the-shelf metal detectors. Two approaches will be considered in parallel: (1) local 3-D imaging and (2) signal analysis. In addition, work associated with soil influences and the optimization of metrological methods will also be carried out (see Figure 1).
Local 3-D imaging is made up of two areas: First, new inversion methods will be developed and tested by the project network enabling inverse problems to be calculated in real time and thus be implemented directly at the demining site. Methods which will be applied include the linear sampling method, the factorization method, the point source method and the approximative inverse. Second, in the field of forward calculation, the aim is to carry out a real-time analysis of detector signals using rapid forward calculation methods. Based on a model assumption for the metal distribution in the ground, the resulting »virtual« detector signals can be calculated and then compared with the real detector signals. The model assumptions are then altered iteratively to enable the best possible match between the data calculated and the data measured to be achieved.
The metal detector signals are also analyzed using signal analysis methods. On one hand, modern feature extraction and classification concepts are used such as support vector machines, neuronal networks and Bayes classifiers. On the other hand, phenomenological or physical correlations are also taken into account in the implementation. Additionally, work concerned with pre-processing data (e.g. noise removal) will be carried out.
In total, ten institutes working in the fields of applied and numerical mathematics, electrical engineering, geophysics and non-destructive testing are participating in the project network. A list of the participating institutes is shown in Figure 2. The project network will run over a total of three years; the individual projects started in October 2003.

: http://publica.fraunhofer.de/dokumente/N-26537.html