Employing Deep Learning to Find Slow Pions in the Pixel Detector in the Belle II Experiment

Sumitted to PubDB: 2021-12-31

Category: Master Thesis, Visibility: Public

Tags:

Authors	Johannes Bilk, Jens Lange
Date	2021-01-01
Belle II Number	BELLE2-MTHESIS-2021-081
Abstract	Slow pions (pT<230 MeV/c) are detected and classified based exclisively upon their cluster shape in the PXD, using a neural network (NN) performing image processing. Input to the NN is a 9x9 pixel matrix. ADC values are encoded as grayscale values. The NN uses convolution to encode neighborhood relations in the pixel data. 78% efficiency ("precision") of correct pion identification and 79% rejection of background is achieved.
Conference	Giessen