SJf_Wettbewerbs_Broschüre_2007 - Die Goldene Sonne am Calanda
SJf_Wettbewerbs_Broschüre_2007 - Die Goldene Sonne am Calanda
SJf_Wettbewerbs_Broschüre_2007 - Die Goldene Sonne am Calanda
Sie wollen auch ein ePaper? Erhöhen Sie die Reichweite Ihrer Titel.
YUMPU macht aus Druck-PDFs automatisch weboptimierte ePaper, die Google liebt.
Spektralanalyse des Elektroenzephalogr<strong>am</strong>ms zur Klassifizierung<br />
durch ein Neuronales Netz<br />
Introduction<br />
Brain waves are electrical activities of the brain which are manifested as alternating potential<br />
differences at the scalp surface and are also known as “electroencephalogr<strong>am</strong>” (EEG). The<br />
alternating potential differences encompass a range of typically ±75 microvolt under the<br />
condition of quiet wakefulness, while body movements, in particular movements of legs and<br />
arms, produce somewhat higher voltages.<br />
Principal goals<br />
In our project we aimed at distinguishing between several human body movements by analysing<br />
and classifying the underlying brain electrical activities.<br />
Methods<br />
This task was tackled with the help of spectral analyses along with neural networks. Neural<br />
network analysis models the human brain through neurons that are organized in layers and<br />
interconnected to each other in a variety of ways. Thus, neural networks are able to process any<br />
kind of input stimuli and to generate clearly identifiable responses to these stimuli.<br />
Data material<br />
Using an experimental design with seven different types of body movements (“experimental<br />
conditions”) along with repeated assessments on the s<strong>am</strong>e individual at weekly intervals, we<br />
were able to quantify brain waves, their within-subject fluctuations and their between-subject<br />
variations. Having extracted the spectral information inherent in EEG time series, we applied<br />
neural network analysis in order to construct classifiers that predicted the seven experimental<br />
conditions from the EEG spectral values.<br />
Results<br />
After a suitable phase of learning, the algorithm yielded a final neural network that classified the<br />
seven experimental conditions at a rate of 90% correctly classified probes.<br />
Discussion<br />
As our project was carried out in the sense of a pilot investigation, repeated assessments on<br />
only two test persons are currently available, so that these results are preliminary and lack<br />
general validity. On the other hand, our results can well be considered as a proof of principle,<br />
thus stimulating further investigations that involve a much larger and more representative s<strong>am</strong>ple<br />
of test persons. The progr<strong>am</strong> package developed within the scope of this project will greatly<br />
facilitate attempts in this direction.<br />
Physik / Technik<br />
Raphael Blaser<br />
8600 Dübendorf<br />
1986<br />
Kantonsschule Glattal Dübendorf<br />
Würdigung<br />
<strong>Die</strong>se interessante und vielseitige Arbeit<br />
erlaubt die Erkennung verschiedener Bedingungen<br />
(zum Beispiel Augenblinzeln)<br />
anhand von Hirnstrommessungen. Dazu<br />
haben Raphael Blaser und Eric Stassen<br />
nicht nur ein Computerprogr<strong>am</strong>m geschrieben,<br />
sondern auch zahlreiche Experimente<br />
durchgeführt. <strong>Die</strong> Arbeit zeichnet<br />
sich durch eine systematische und<br />
wissenschaftliche Herangehensweise<br />
aus. Der Bericht ist anschaulich und gut<br />
verständlich.<br />
Prädikat<br />
Hervorragend<br />
Sonderpreis<br />
„European Union Contest for young<br />
Scientists“ in Valencia<br />
Sonderanerkennung<br />
Metrohm Stiftung Herisau<br />
Experte<br />
Dr. Volker Koch<br />
AWK Group AG Zürich, Consultant<br />
45<br />
Eric Stassen<br />
8600 Dübendorf<br />
1987