Breaking the U: Asymmetric U-Net for Object Recognition in Muon Tomography☆

We utilize a U2 net-inspired deep learning model for object detection within a scattering muon tomography setup. Capturing muon readings from scintillators situated above and beneath the examination area, we discern the angular disparities between inbound and outbound muon paths. These angular measurements serve as inputs for a neural network, proficient in forecasting the contours of objects and, to an extent, their constituent materials within the research zone. Our model handles the inherent complexity of the input data, where the angular measurements map onto a 200 x 200 space, while producing an output image of a smaller 40 x 40 resolution, resulting in an asymmetric U-Net architecture that diverges from conventional semantic segmentation models which typically maintain the same input and output dimensions.