You are here: Homepage NKS Reports View document
|NKS Programme Area:
|Principle Component Analysis as Applied to Qualitative Analysis of Mobile Measurement and Monitoring Data Sets (PRICOMOB)
|M. Dowdall, Tero Karhunen, Ellinoora Vikman, Mats Eriksson, Gísli Jónsson,
|Mobile measurement systems are the backbone of most responses to cases of orphan sources. Conducting mobile measurement surveys, irrespective of the platform utilised, is a non-trivial task with respect to the nature of the data being accrued – large volumes of discrete, often highly variable, data points where the signal of interest may be weak, superimposed on a constantly fluctuating background and only present for a tiny proportion of the overall data set. Principal Component Analysis (PCA), one of the most popular multivariate statistical technique, is a flexible statistical procedure that allows for the summarizing of the information content in large data tables by means of a smaller set of “summary indices” that can be more easily visualized and analyzed in order to observe trends, jumps, clusters and outliers. The PRICOMOB project focussed on the application of PCA to mobile measurement and stationary scanning data to assess its performance in identifying source signals from a number of isotopes superimposed on a variable background signal typical of mobile measurement data. The PCA method implemented proved itself a viable method to detect anomalies in spectral time series. A disadvantage of the method employed is that a training data set is needed containing all the features and behavior that are not due to artificial radioactivity. Alternative ways to form the residuals used in deciding whether a measurement contains features not previously seen included the Mahalanobis distance and a modified Euclidean distance. The modified Euclidean distance seemed to result in improved sensitivity for radionuclides that produce peaks, but reduced sensitivity for sources that produce continuums (such as x-rays).
|Principal Component Analysis, Gamma spectrometry, Mobile measurement, time series
|23 Jan 2024
|Number of downloads: