Highlight

Seismic event classification and detection have been important research topics because of their significance and wide applications on hazard assessment and global security. In the real world, seismic data acquisition are always impacted by unavoidable nature factors, which will introduce low-frequency noise to the seismic events of interests. Pre-processing of seismic signal using denoising techniques can be critical to the detection of the seismic events. In our work, we develop an end-to-end framework which can automatically learn the hyper-parameter in the denoising algorithm so that we do not need to manually set the hyper-parameter. Specifically, our network structure consists of two modules, an adaptive filtering module for signal denoising, and a classification module for signal classification. We further develop two mechanisms of the adaptive filtering module, namely, sample-specific mechanism and dataset-specific mechanism.

The full paper can be found on IEEE ICASSP:

Adaptive Filtering for Event Recognition from Noisy Signal: an Application to Earthquake Detection

"Why" adaptive Filtering?

Seismologists usually employ a preprocessing step to denoise the seismic signal by applying a high-pass frequency filter. The selection of the cutoff threshold for high-pass filter can be subjective. In addition, it can be challenging to find the best threshold to denoise signal and maintain as much useful information as possible at the meanwhile. We, on the other hand, design the adaptive filtering module to guide the neural network to automatically select the best cutoff threshold value.

Results

Below we provide some results illustrating adaptive filter when applying to field seismic data.