Data classification and parameter estimations with deep learning to the simulated time-domain high-frequency gravitational waves detections

High-frequency gravitational wave (HFGW) detection is a great challenge, as its signal is significantly weak compared with the relevant background noise in the same frequency bands. Therefore, besides designing and running the feasible installation for the experimental weak-signal detection, developing various effective approaches to process the big detected data for extracting the information about the GWs is also particularly important. In this paper, we focus on the simulated time-domain detected data of the electromagnetic response of the GWs in high-frequency band, typically such as Gigahertz. Specifically, we develop an effective deep learning method to implement the classification of the simulated detection data, which includes the strong electromagnetic background noise in the same frequency band, for the parameter estimations of the HFGWs. The simulatively detected data is generated by the transverse first-order electromagnetic responses of the HFGWs passing through a high stationary magnetic field biased by a high-frequency Gaussian beam. We propose a convolutional neural network model to implement the classification of the simulated detection data, whose accuracy can reach more than 90%. With these data being served as the positive sample datasets, the physical parameters of the simulatively detected HFGWs can be effectively estimated by matching the sample datasets with the noise-free template library one by one. The confidence levels of these extracted parameters can reach 95% in the corresponding confidence interval. Through the multiple data experiments, the effectiveness and reliability of the proposed data processing method are verified. The proposed method could be generalized to big data processing for the detection of experimental HFGWs in the future.