用全卷积网络进行多频点源探测：真实微波天空模拟中的性能
Multi-frequency point source detection with fully convolutional networks: Performance in realistic microwave sky simulations
摘要：
点源探测是未来宇宙微波背景实验中的一个重要问题，因为它是影响小尺度微波背景信号恢复的主要因素之一。改进其多频检测将允许考虑在使用逐信道方法提取PS时被忽略的有价值的信息。我们开发了一种基于神经网络（NN）的方法来检测PS在多频现实模拟和比较其性能对最流行的方法之一，矩阵滤波器。使用的频率是143，217和353 GHz，我们施加了30度的银河切割。我们通过在PS图中添加污染信号来产生模拟，这些污染信号包括CMB、宇宙红外背景、银河系热辐射、热Sunyaev-Zel'dovich效应和仪器噪声。这些模拟用于训练两个NN，称为平面和光谱MultiPoSeIDoN。第一个考虑具有平坦光谱的PS，第二个更现实，因为它考虑了PS的光谱行为。使用60 mJy的检测限，Flat MultiPoSeIDoN在143，217和353 GHz的光谱情况下分别在58 mJy和79，71和60达到90%的完整性水平，而矩阵滤波器在84，79和123 mJy达到它。使用更安全的4σ检测限无助于改善这些结果。在所有情况下，MultiPoSeIDoN获得比过滤器少得多的虚假源。在100 mJy以上的情况下，神经网络以10%的相对误差恢复检测的通量密度，而在150 mJy以上的情况下，滤波器恢复检测的通量密度。基于这些结果，在未来的CMB实验中，神经网络是替代滤波器检测多频PS的理想候选者。此外，我们已经表明，多频方法可以检测源，比单频方法更高的精度也基于神经网络。
英文摘要：
Point Source (PS) detection is an important issue for future Cosmic Microwave Background (CMB) experiments since they are one of the main contaminants to the recovery of CMB signal at small scales. Improving its multifrequency detection would allow to take into account valuable information otherwise neglected when extracting PS using a channel-by-channel approach. We develop a method based on Neural Networks (NNs) to detect PS in multifrequency realistic simulations and compare its performance against one of the most popular methods, the matrix filters. The frequencies used are 143, 217 and 353 GHz and we impose a Galactic cut of 30 degrees. We produce simulations by adding contaminating signals to the PS maps as the CMB, the Cosmic Infrared Background, the Galactic thermal emission, the thermal Sunyaev-Zel'dovich effect and the instrumental noise. These simulations are used to train two NNs called Flat and Spectral MultiPoSeIDoN. The first one considers PS with a flat spectrum and the second one is more realistic because it takes into account the spectral behavior of the PS. Using a detection limit of 60 mJy, Flat MultiPoSeIDoN reachs the 90% of completeness level at 58 mJy and at 79, 71 and 60 for the spectral case at 143, 217 and 353 GHz respectively, while the matrix filters reach it at 84, 79 and 123 mJy. Using safer 4σ detection limit does not help to improve these results. In all cases, MultiPoSeIDoN obtain a much lower number of spurious sources than the filter. The NNs recover the flux density of the detections with a relative error of 10% above 100 mJy, while the filter above 150 mJy. Based on the results, NNs are the perfect candidates to substitute filters to detect multifrequency PS in future CMB experiments. Moreover, we have shown that a multifrequency approach can detect sources with higher accuracy than single-frequency approaches also based on NNs.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
