Enhancing Void and Contour Image Quality in Muon Tomography Using GEANT4 Simulations and ResUNet Architecture Mutia Delina(a)*, Parulian Renaldi(a), and Janaka Adassuriya (b)
(a)Physics Department, Faculty of Mathematics and Natural Sciences, Universitas Negeri Jakarta, Jl. Rawamangun Muka, Jakarta Timur 13220, Indonesia
*mutia_delina[at]unj.ac.id
(b)Astronomy and Space Science Unit, Department of Physics, University of Colombo, Sri Lanka
Abstract
This study proposed to enhance the low-quality profile produced by short-duration muon tomography image processes by conducting image segmentation and masking with Residual UNet Architecture. Muon tomography images the void and contour of an object once a muon cosmic ray passes through the object. Unfortunately, muon tomography still has several disadvantages today, such as a big image needs a long duration of image process. Meanwhile, a short-duration image process will generate a low- quality profile. In the process, the dataset was generated from GEANT4. The procedures were annotating the image using a brush to get perfect pixels, resizing the images from 112 pixels to 256 pixels, training the image with Kaggle Notebook, predicting the results, and then comparing the prediction results to existing ground truth with a confusion matrix. The comparison results showed that the prediction image accuracy with Residual UNet Architecture was 99,7%, close to the ground truth image. Therefore, the Residual UNet Architecture has successfully generated high-accuracy tomography image results.
