2021/10/01 by Łukasz Bibrzycki, Bibrzycki, Łukasz, David Alvarez-Castillo +23
Computer Science · Physics and Astronomy · #Astrophysics and Cosmic Phenomena #Computational Physics and Python Applications #FOS: Electrical engineering #FOS: Physical sciences #Gamma-ray bursts and supernovae #Instrumentation and Detectors (physics.ins-det) #Particle Detector Development and Performance #Signal Processing (eess.SP) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2110.00297
openalex publication_date 2021/10/01 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
The wealth of smartphone data collected by the Cosmic Ray Extremely\nDistributed Observatory(CREDO) greatly surpasses the capabilities of manual\nanalysis. So, efficient means of rejectingthe non-cosmic-ray noise and\nidentification of signals attributable to extensive air showers arenecessary.\nTo address these problems we discuss a Convolutional Neural Network-based\nmethod ofartefact rejection and complementary method of particle identification\nbased on common statisticalclassifiers as well as their ensemble extensions.\nThese approaches are based on supervised learning,so we need to provide a\nrepresentative subset of the CREDO dataset for training and\nvalidation.According to this approach over 2300 images were chosen and manually\nlabeled by 5 judges.The images were split into spot, track, worm (collectively\nnamed signals) and artefact classes.Then the preprocessing consisting of\nluminance summation of RGB channels (grayscaling) andbackground removal by\nadaptive thresholding was performed. For purposes of artefact rejectionthe\nbinary CNN-based classifier was proposed which is able to distinguish between\nartefacts andsignals. The classifier was fed with input data in the form of\nDaubechies wavelet transformedimages. In the case of cosmic ray signal\nclassification, the well-known feature-based classifierswere considered. As\nfeature descriptors, we used Zernike moments with additional feature relatedto\ntotal image luminance. For the problem of artefact rejection, we obtained an\naccuracy of 99%. For the 4-class signal classification, the best performing\nclassifiers achieved a recognition rate of 88%.\n