Exploring novel machine learning applications in muography: A promising frontier

The utilisation of cosmic ray muons in muon radiography provides a non-invasive imaging technique for the visualisation of large, dense structures without the necessity for particle accelerators or radioactive sources. This study introduces a neural network-based approach for the automatic detection of anomalies within a muon radiography, with a specific focus on the identification of cavities in mining environments. The employment of the UNet architecture has enabled the training of a model on simulated datasets, resulting in the achievement of high precision in anomaly segmentation. The validation of the model using real muon radiographic data from the Temperino mine has confirmed its effectiveness. Our findings indicate that integrating neural networks with muon radiography enhances anomaly detection efficiency, with significant potential for future advancements.