Poster Abstract

P3.15 Yunfei Xu (National Astronomical Observatories, Chinese Academy of Sciences)

Theme: Data science challenges: tools from statistics to machine learning

X-Ray Source Detection of Einstein Probe Satellite Data based on Deep Learning

Einstein Probe (EP) satellite will be launched in 2022, it is dedicated to the discovery and exploration of X-ray transients and explosive celestial bodies in the universe. The satellite contains a wide field of view X-ray telescope (WXT) for all-sky monitoring. The observation field of WXT is extremely large, exceeding 3,600 square degrees, which is an efficient tool for searching for the all-sky X-ray transients.

Due to WXT’s structural characteristics, each x-ray source in WXT's image has a cross shape and broken dark lines appear on the cross arm, this makes the traditional source detection method not work well. In order to solve this problem, this study used the deep learning method to carry out the EP simulated data X-ray source detection.

Based on the point spread function and background noise information of WXT, 30,000 simulated X-ray sources with different brightness and positions are randomly generated. This data was input into our deep learning network as the training dataset, a modified ResNet was applied as the X-ray source recognition framework, which can also output the estimated brightness of each source.

By comparing the detection results with a test dataset, our method has a strong ability to distinguish between the foreground(source pixels) and the background(noise), can determine the location of X-ray transients, and also can correctly detect the darkest source of the EP observation capability (limit flow 0.3mCrab), although the accuracy of source brightness evaluation needs to be improved (67.62%), the results can meet the needs of EP observation source detection.