Seismic Signal Clustering Using Deep-Self-Supervised Networks

Date: 4/24/2019

Time: 11:15 AM

Room: Elliott Bay

We present a method for unsupervised clustering of seismic signals using deep neural networks. In this approach, the clustering is performed in the feature space and the feature learning and clustering are optimized simultaneously resulting in learning feature representations that are more suitable for specific clustering tasks. The training process is done in a self-supervised fashion where the targets are generated from the input data. To demonstrate the application of this method for seismic signal processing, we designed two different networks mainly consist of full convolution and pooling layers for discriminating the waveforms with different polarities of first motions and hypocentral distances. Our method resulted in precisions close to those of supervised methods but without a need for labeled data and using much smaller datasets.

Presenting Author: Mostafa Mousavi

Authors