Identification and quantification in mixtures of multisensor signals

In any system, what is observed is in principle a sum of the signals of the pure sources as a function of their relative part of the mixture. When the pure sources are known, or have been estimated, the objective is often to identify pure sources in new observations. The typical approach is to compare a new observation to a library of possible pure signals or to calculate the Mahalanobis distance, but this approach does not quantify mixtures and is not the best approach for unique identification when the sources are highly correlated.

The suggested method takes into account a number of criteria such as partial correlation, peak finding and various distance measures. A proof of concept study has already been conducted with promising results. Extension of this work could be to add noise for robustness studies and implement a moving window option to identify which sources that are present in a dynamic systes. The work also comprises an investigation of various methods for estimating pure soures such as Independent Component Analysis (ICA), Multivariate CurveResolution (MCR) and combinations of these. Typical applications are found within spectroscopy, acoustics and medical multisensor instruments.

Dette er i første omgang en prosjektoppgave, hvor studenten vil kunne ta tak i en begrenset del av problemstillingen. Oppgaven vil kunne utvides til en masteroppgave.

Det anbefales å ta temaet TTK19 Strukturer og sammenhenger i komplekse systemer (3.75 studiepoeng), se

Oppgaven er knyttet til NTNUs satsning innen Big Data Cybernetics.

Det vil være begrensninger på opphavsrettigheter, herunder publisering.

Hovedveileder: Prof. Harald Martens (NTNU ITK / Idletechs)
Medveileder: Dr. Frank Westad (Camo AS)

Tilbake til menyen over mulige masteroppgaver