Speaker
Description
Transverse beam imaging in radiation areas can be supported by relaying scintillation light through a multimode fiber (MMF) to a camera placed in a shielded area. However, the MMF scrambles the input, so a trained model is required to recover the beam distribution. This work studies a data efficient calibration method in which measured input and MMF output basis pairs are used as building blocks to synthesize training data for the reconstruction model. After an initial digital micromirror device based validation, the method was assessed using real beam data from CERN CLEAR, where data synthesized from a raster scan basis were used to train a convolutional autoencoder. The best model using this strategy achieved 7.37% mean normalized root mean square error (RMSE) across four transverse beam parameters, compared with 6.02% for a random scan reference model using roughly twice as many fully paired random scan samples. These results suggest that basis-based synthesis training, when combined with suitable beam image priors, can reduce reliance on large random scan MMF calibration datasets by replacing part of the calibration with a controlled scan of fixed size.
Funding Agency
This work was supported by the Science and Technology Facilities Council (STFC) through the LIV.INNO Centre for Doctoral Training under grant agreement ST/W006766/1 and CERN.
| In which format do you inted to submit your paper? | LaTeX |
|---|---|
| Preprint marking on your proceeding paper | I wish my paper to be marked as preprint. |