Gravitational Wave Signal Extraction Based On Convolutional Neural Networks

Ten binary black hole(BBH)coalescences were detected during the first(O1)and second(O2)observing runs of LIGO/Virgo,which has led us to the era of gravitationalwave(GW)astronomy.LIGO/Virgo third(O3)observing run has already begun,and more gravitational-wave events are expected to be observed.Developing fast and robust analysis methods to extract gravitational-wave signals from data will be urgent as the improvement of detectors' sensitivities in the upcoming years.Currently,the standard method for detecting transient gravitational-wave signals is the matched filtering technique.Although it works very well in the extraction of weak signals,a huge amount of computational cost is required to process the data which might contain the gravitational-wave signals.This makes the matched filtering technique unsuitable for the future multi-messenger observations.Gabbard et al.have demonstrated that the results of matched filtering analysis in Gaussian noise can be closely reproduced,when deep learning is applied to gravitational wave time series data.And convolutional neural networks can achieve the sensitivity of matched filtering in the recognization of the gravitational-wave signals with high efficiency [Phys.Rev.Lett.,120,141103(2018)].In the process of implementing the convolutional neural network model,we found that the classification effect of the model on the training set is cool,but the classification effect on the test set is poor,that is,the model has an overfitting problem.The paper will optimize the model of Gabbard et al.to improve accuracy.The convolutional neural networks typically have alternating convolutional layers and max pooling layers,followed by a small number of fully connected layers.We set the stride of the max pooling layer and the corresponding convolutional layer to 2,followed by a dropout layer to alleviate overfitting in the original model.We used the optimized model and the original model on the same data set for various tests,and plotted the Receiver Operating Characteristic curve,referred to as the ROC curve.And the area under the ROC curve was calculated.We find that the optimized model can increase the area under the ROC curve between 0.01924 to 0.04800 without increasing the running time.In addition,in order to verify the robustness of the optimization model to noise,the paper changed the amplitude of the Gaussian noise,added the changed noise to the gravitational-wave time series data to form a new data set,and used the optimization model and the original model for the new testing on the data set,it also proves that compared with the original model,the area under ROC curve of the optimized model will still increase.