Use of a UNet network for the identification of cavities inside mines

The utilisation of cosmic ray muons in muon radiography provides a non-invasive methodology for imaging large and dense structures without the necessity for particle accelerators or radioactive materials. This research presents a neural network-driven approach for the automatic identification of anomalies in muon radiographs, with a particular emphasis on detecting cavities in mining settings. The UNet architecture was employed to train a model on simulated data, resulting in high accuracy in anomaly segmentation. The model's effectiveness was validated using actual muon radiographic data from the Temperino mine. The results demonstrate that combining neural networks with muon radiography improves anomaly detection efficiency and holds considerable promise for future developments.