Denoising of Seismic Signals Recorded at Local to Near-Regional Distances Using Deep Convolutional Neural Networks

Session: Leveraging Advanced Detection, Association and Source Characterization in Network Seismology [Poster]
Type: Poster
Date: 4/30/2020
Time: 08:00 AM
Room: Ballroom
Description: 

Seismic waveform data are generally contaminated by noise from various sources. Suppressing this noise effectively, so that the remaining signal of interest can be successfully exploited, remains a fundamental problem for the seismological community. Among available noise suppression methods, the most commonly used approaches are those based on frequency filtering. These methods, however, are less effective when the signal of interest and noise share similar frequency bands. Also, these methods are specific to the type of noise, and must be tuned on a case-by-case basis. Inspired by recent work in seismology (Zhu et al., 2018) and source separation studies in the field of Music Information Retrieval, we implemented a seismic signal denoising method that uses a trained deep convolutional neural network (DNN) model to decompose an input waveform into signal of intertest and noise. In our approach, the DNN provides a signal mask and a noise mask for an input signal. The Short-Time Fourier Transform (STFT) of the estimated signal is obtained by multiplying the signal mask with the STFT of the input signal. To build and test the denoiser, we used carefully compiled signal and noise datasets of seismograms recorded by the University of Utah Seismograph Station network. The signal dataset consists of 3,188 high-quality waveforms recorded at station BRPU that were generated by earthquakes at distances of about 10-670 km. The noise dataset contains 15,426 waveforms from various noise sources and various stations. The two datasets were each randomly divided into training, validation, and test sets. For each set, noisy waveforms were formed by summing each signal waveform and a randomly selected noise waveform. Preliminary results of the denoiser test runs are promising, with the recovered signals having high similarity and high source-to-distortion ratios with respect to the target signals.

Presenting Author: Rigobert Tibi

Authors