论文标题
使用深度学习从二进制黑洞中检测爱因斯坦望远镜引力波信号
Detection of Einstein Telescope gravitational wave signals from binary black holes using deep learning
论文作者
论文摘要
来自第三代重力波（GWS）爱因斯坦望远镜（ET）检测器的预期数据量将使传统的GWS搜索方法（例如匹配过滤）不切实际。这是由于需要大的模板库以及波形建模的困难。相比之下，机器学习（ML）算法显示了GWS数据分析的有希望的替代方法，其中ML可用于开发半自动和自动的工具，用于检测GWS源的检测，降解和参数估计。与第二代探测器相比，ET将具有更宽的可访问频带，但也将具有较低的噪声。 ET分别针对每年1E5-1E6和每年7E4的二元黑洞（BBH）和二元中子星（BNSS）的检测率分别具有检测率。在这项工作中，我们探讨了使用卷积神经网络（CNN）检测埋在高斯噪声中的合成GWS信号中的BBHS合并的可能性和效率。数据是使用开源工具根据ETS参数生成的。在不执行数据美白或应用带通滤波的情况下，我们培训了四个CNN网络，具有计算机视觉中最先进的性能，即VGG，Resnet和Densenet。 Resnet的性能明显更好，可检测SNR为8或更高的BBHS源，精度为98.5％，而SNR范围分别为7-8、6-7、5-6和4-5的源的SNR精度为92.5％，85％，60％和62％。在定性评估中，Resnet能够以4.3 SNR检测到60 GPC的BBHS合并。还表明，在长期序列数据上使用CNN进行BBHS合并在计算上是有效的，可用于近实时检测。
The expected volume of data from the third-generation gravitational waves (GWs) Einstein Telescope (ET) detector would make traditional GWs search methods such as match filtering impractical. This is due to the large template bank required and the difficulties in waveforms modelling. In contrast, machine learning (ML) algorithms have shown a promising alternative for GWs data analysis, where ML can be used in developing semi-automatic and automatic tools for the detection, denoising and parameter estimation of GWs sources. Compared to second generation detectors, ET will have a wider accessible frequency band but also a lower noise. The ET will have a detection rate for Binary Black Holes (BBHs) and Binary Neutron Stars (BNSs) of order 1e5 - 1e6 per year and 7e4 per year respectively. In this work, we explore the possibility and efficiency of using convolutional neural networks (CNNs) for the detection of BBHs mergers in synthetic GWs signals buried in gaussian noise. The data was generated according to the ETs parameters using open-source tools. Without performing data whitening or applying bandpass filtering, we trained four CNN networks with the state-of-the-art performance in computer vision, namely VGG, ResNet and DenseNet. ResNet has significantly better performance, detecting BBHs sources with SNR of 8 or higher with 98.5% accuracy, and with 92.5%, 85%, 60% and 62% accuracy for sources with SNR range of 7-8, 6-7, 5-6 and 4-5 respectively. ResNet, in qualitative evaluation, was able to detect a BBHs merger at 60 Gpc with 4.3 SNR. It was also shown that, using CNN for BBHs merger on long time series data is computationally efficient, and can be used for near-real-time detection.
