Tomographic Reconstruction of Imaging Diagnostics with a Generative Adversarial Network

We have developed a tomographic reconstruction method using a conditional Generative Adversarial Network to obtain local-intensity profiles from imaging-diagnostic data. To train the network we prepared pairs of local-emissivity and line-integrated images that simulate the experimental system. After validating the accuracy of the trained network, we used it to reconstruct a local image from a measured line-integrated image. We applied this procedure to the He II-emission imaging diagnostic for RT-1 magnetospheric plasmas, including the effects of stray light within the measured image to remove reflections from the chamber walls in the reconstruction. The local intensity profiles we obtain clearly elucidate the effect of ion-cyclotron-resonance heating. This method is a powerful tool for systems where it is difficult to solve the inversion problem due to the involved contributions of nonlocal optical effects or measurement restrictions.
source:https://doi.org/10.1585/pfr.14.1202117
identifier:0000-0003-1088-8237