Mainz 2026 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
A: Fachverband Atomphysik
A 15: Poster – Highly Charged Ions and their Applications
A 15.5: Poster
Dienstag, 3. März 2026, 17:00–19:00, Philo 1. OG
Convolutional Neural Network for fast and accurate ion-number counting with Micro-Channel Plate detector — •Jun Huang1, Stefan Ringleb1, Manuel Vogel2, and Thomas Stöhlker1,2,3 — 1Friedrich-Schiller-Universität Jena — 2GSI Helmholtzzentrum für Schwerionenforschung Darmstadt — 3Helmholtz-Institut Jena
Micro-channel plates (MCP) are ideal detectors for detecting and counting ions extracted from ion traps as long as the ion rate is low enough to detect single ion hits. During our laser experiment with the HILITE Penning trap, a huge amount of ions are expected to arrive at the MCP within microseconds, hence overlapping of multiple ions is very probable. The manual data evaluation is time consuming, as a large amount of datasets is expected. A Convolutional Neural Network (CNN) is extremely helpful in handling two-ion signals overlapping in order to recognise and count them rapidly in large datasets. In our evaluation method, single-ion signals are first hand selected and trained into a CNN of Single Ion Model (SIM) for detection with high accuracy. Due to the low number of existing double-ion data for model training, the single-ion data is used to create artificial double-ion signal, which enrich the training data for CNN of Double Ion Model (DIM). In our created CNN models, single-ion signals and double-ion signals are rapidly recognised by SIM and DIM, and the detection is then used to obtain the total ion number. We will present the model and its ion counting capabilities with a special focus on detection accuracy.
Keywords: neural network; penning trap; data analysis