Super-resolution for x-ray applications with pixelated semiconductor tracking detectors using convolutional neural networks
In a semiconductor tracking detector, a single X-ray photon can create signals in a cluster of adjacent pixels. We present a novel technique to reconstruct the points of entry (PoEs) of X-ray photons from these clusters based on a convolutional neural network (CNN). The new method allows improving the spatial resolution into subpixel regime. Beside the improved accuracy of the reconstruction, the method is much less computational intensive than conventional event analyses and therefore can be run even on less powerful machines in realtime. Due to its special architecture, the CNN can handle different frame sizes without adjustments or retraining processes.
Paper
Full text
Super-resolution for x-ray applications with pixelated semiconductor tracking detectors using convolutional neural networks
Semantic Scholar · Physics · 2020
Abstract
In a semiconductor tracking detector, a single X-ray photon can create signals in a cluster of adjacent pixels. We present a novel technique to reconstruct the points of entry (PoEs) of X-ray photons from these clusters based on a convolutional neural network (CNN). The new method allows improving the spatial resolution into subpixel regime. Beside the improved accuracy of the reconstruction, the method is much less computational intensive than conventional event analyses and therefore can be run even on less powerful machines in realtime. Due to its special architecture, the CNN can handle different frame sizes without adjustments or retraining processes.