Recovering CMB Polarization Signals with Machine Learning 后印本

摘要: Primordial B-mode detection is one of the main goals of the current and future CMB experiments. However, the weak B-mode signal is overshadowed by several Galactic polarized emissions, such as the thermal dust emission and the synchrotron radiation. Subtracting the foreground components from CMB observations is one of the key challenges in the search for primordial B-mode signal. Here, we construct a deep convolutional neural network (CNN) model, called CMBFSCNN (Cosmic Microwave Background Foreground Subtraction with CNN), which can cleanly remove various foreground components from the simulated CMB observational maps with the sensitivity of the CMB-S4 experiment. The noisy CMB Q (or U) maps are recovered with a mean absolute difference of $0.018 \pm 0.023\ \mu$K (or $0.021 \pm 0.028\ \mu$K). To remove residual instrumental noise in the foreground-cleaned map, inspired by the Needlet Internal Linear Combination method, we divide the whole data into two ``half-split maps'' which share the same sky signal but with uncorrelated noise, and perform the cross-correlation technique to reduce the instrumental noise effect at the power spectrum level. We find that the CMB EE and BB power spectra can be precisely recovered with significantly reduced noise effects. Finally, we apply this pipeline on the current Planck observations. As expected, various foregrounds have been cleanly removed on the Planck observational maps and the recovered EE and BB power spectra are in good agreement with the Planck official results.