Dresden 2020 – wissenschaftliches Programm

Die DPG-Frühjahrstagung in Dresden musste abgesagt werden! Lesen Sie mehr ...

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

KFM: Fachverband Kristalline Festkörper und deren Mikrostruktur

KFM 15: Postersession KFM

KFM 15.22: Poster

Donnerstag, 19. März 2020, 16:00–18:30, P2/1OG

A supervised Machine Learning Approach for Shape sensitive Detector Pulse Discrimination in Positron Spectroscopy Applications — Danny Petschke and •Torsten Staab — Department of Chemistry and Pharmacy, University of Wuerzburg, Germany

The acquisition of high-quality positron spectra is crucial for a profound analysis, i.e. the correct decomposition to obtain the true parameters. Since the introduction of digital spectrometers for the techniques of PALS and CDBS, this is generally achieved by applying various physical filters on the digitized output-pulses from PMTs or HPGe-detectors prior to spectra generation. For instance, pile-up events can be easily rejected by applying pulse area or shape sensitive filters, which significantly increases the peak-to-background ratio.

Here, we present a novel approach for shape-sensitive discrimination of detector outputpulses using supervised machine learning (ML) based on a simple probabilistic classification model: the naive Gaussian Bayes classifier. In general, naive Bayes methods find wide application for many real-world problems such as famously applied for email spam filtering or document classification. Their algorithms are relatively simple to implement and, moreover, perform extremely fast compared to more sophisticated methods in training and predicting on high-dimensional datasets, e.g. detector-output pulses. We compared the quality and decomposability of lifetime spectra acquired on pure iron from a single measurement (pulse stream): (1) generated by applying the ML approach to lifetime spectra generated using exclusively physically filtering.