Auto-windowed Super-virtual Interferometry via Machine Learning: A Strategy of First-arrival Traveltime Automatic Picking for Noisy Seismic Data

Supervirtual interferometry (SVI) was developed to significantly enhance the signal-to-noise ratio of noisy first arrivals. However, a time window must be specified that contains these first arrivals, and the window should be no wider than several times the dominant period of the source wavelet. The accurate specification of this window is very challenging for noisy data and involves manual picking. To overcome this problem, we propose to automatically pick these windows via machine learning methods. Convolutional neural network (CNN) and density-based spatial clustering of applications with noise (DBSCAN) are used to distinguish first-arrival signals completely buried in noise. Numerical tests validate that this method can accurately specify the correct window as well as that of a human interpreter. The benefit is an automatic means for picking first-arrival traveltimes in noisy traces from a large 3D data set.