Speaker
Description
The Cosmic-Ray Extremely Distributed Observatory (CREDO) uses a geographically distributed network of smartphones to detect secondary cosmic radiation. A major challenge in such a citizen-science experiment is the heterogeneous quality of data and the large number of artefacts caused by light leakage, damaged pixels and improper detector operation. This poster presents a data-processing framework developed to prepare CREDO measurements for statistical anomaly searches. The procedure combines rule-based image filters with a convolutional neural network trained on 10,000 manually classified images of particle-like tracks and artefacts. The model achieved 93.66% validation accuracy and an AUC of 0.979. After data cleaning, detector activity was analysed individually using exposure-corrected five-minute time windows and Poisson statistics. This work constitutes the first systematic analysis of this CREDO dataset using such a combined pipeline. Reliable preprocessing, including AI-based classification, is a necessary step before investigating possible correlations between secondary cosmic-ray detections and other phenomena, such as seismic activity.