Recognition and classification of the cosmic-ray events in images captured by CMOS/CCD cameras

TL;DR

This paper introduces a machine learning-based method for recognizing and classifying cosmic-ray events in images captured by various CMOS/CCD cameras, utilizing shape analysis and metadata for improved detection.

Contribution

The paper presents a novel approach combining image shape analysis and metadata to classify cosmic-ray events, including muons and electrons, in diverse camera images.

Findings

Successfully classified cosmic-ray events with shape and metadata analysis.

Estimated azimuth of muon tracks in images.

Analyzed over 2 million images from global sources.

Abstract

Muons and other ionizing radiation produced by cosmic rays and radiative decays affect CMOS/CCD sensor. When particles colliding with sensors atoms cause specific kind of noise on images recorded by cameras. We present a concept and preliminary implementation of method for recognizing those events and algorithms for image processing and their classification by machine learning. Our method consists of analyzing the shape of traces present in images recorded by a camera sensor and metadata related to an image like camera model, GPS location of camera, vertical and horizontal orientation of a camera sensor, timestamp of image acquisition, and other events recognized near-by sensors. The so created feature vectors are classified as either a muon-like event, an electron-like event or the other event, possibly noise. For muon-like events our method estimates azimuth of a muon track. Source of the data is database of CREDO (Cosmic-Ray Extremely Distributed Observatory) project and ESO (European Southern Observatory) archives. The telescope dark frames from ESO are analysed. CREDO project collected so far over 2 millions images of events from many kinds of cameralike: smartphones camera, laptop webcams and Internet of Things cameras localised around the globe.