CMB Signal Extraction In Sky Map Based On Deep Learning

Cosmic microwave background radiation(CMB)is the afterglow of the Big Bang.It is the physical limit of human understanding of the early universe.In addition,CMB is affected by the late evolution of the universe in the process of propagation,and these effects will leave traces on the CMB.Therefore,CMB photons are one of the most important observational signals in astronomy.They are important for studying the origin and development of the universe,and deepen the understanding of the composition of the universe,the composition of matter and a series of basic physical processes.In the process of CMB observation,it will be polluted by various signals.For the first time,deep learning is applied to remove the foreground pollution of CMB,and the temperature fluctuation map and polarization map of CMB are deeply studied.For the CMB temperature fluctuation graph,the depth residual self-encoder model is used.Compared with the traditional astronomical method NILC,the mean square error is reduced by 98.29%and the processing speed is increased by 10.18 times.The CMB polarization map is divided into two polarization directions of Q and U.This paper proposes a cross-layer connection residual self-encoder model based on the depth residual self-encoder.Compared with the traditional astronomical method,the Q-direction polarization map is square.The error is reduced by 99.17%,and the mean square error of the polarization diagram in the U direction is reduced by 99.71%,and an almost completely pure polarization map is obtained.