Speaker
Description
Automatic image reconstruction tools are essential in fields such as physics, astronomy, and biology. In particle accelerators like CERN’s LHC, they are particularly important for evaluating beam quality through beam distribution measurement as position, profile, and emittance. Traditional analysis tools no longer meet the accuracy and efficiency demands of future facilities. We developed a new AI/DL-based digital tool for 2D transverse phase-space distributions and scanner image denoising aimed at improving the accuracy of RMS emittance measurements. Our focus was to enhance beam halo characterization, ultimately contributing to reduce transport losses, and enabling more sustainable accelerator operations.
Challenges are related to noisy experimental data, lack of ground-truth reference images, noise model, and limited training datasets. Our unsupervised deep convolutional neural network (DCNN) can denoise a single image using only itself as input thus greatly improving the accuracy of beam surface area estimation and hence the RMS emittance measurement.
| Paper status | Proceeding files received |
|---|