Optimization of deep learning method on track reconstruction for X-ray polarimetry with gas pixel detectors

The reconstruction of the photoelectron tracks in X-ray polarimetric detectors based on Gas Pixel Detectors (GPD) is crucial for polarization detection. In addition to traditional moment analysis methods, the convolutional neural network (CNN) is also a noteworthy approach. However, most existing CNN methods for polarization detection have only been effectively validated on simulated data, and the few methods validated on experimental data have not yielded satisfactory results. We have improved the CNN algorithm for reconstructing the emission direction of photoelectron tracks in X-ray polarimetric detectors. We tested this algorithm using calibration data from the detectors of the PolarLight mission and the Polarimetry Focusing Array (PFA) onboard the enhanced X-ray Timing and Polarimetry (eXTP) mission. The results indicate that the optimized deep learning model increased the modulation factor by approximately 0.02 over the 2-8 keV energy range and only introduced a small systematic error. This can enhance the sensitivity of polarization detector in the low-energy range. Additionally, the computational resources required for the model are much lower than the previous CNN models.