GlitchNet: A Glitch Detection and Removal System for SEIS Records Based on Deep Learning
Wuchuan Xu, Qiwen Zhu, Li Zhao
Abstract
Abstract We have developed a system based on deep learning for the detection and removal of glitches, a special type of noise that is common in the continuous data recorded by the Seismic Experiment for Interior Structure (SEIS) system deployed on Mars during the InSight mission. We first used the existing algorithms to build datasets of glitches and noises that are used to train the detection and removal networks. Then glitch detection was realized by a five-layer convolutional neural network (CNN); glitch removal is fulfilled by subtracting from the raw record a glitch waveform constructed using a deep autoencoder network. The resulting GlitchNet, a combination of our CNN and autoencoder network, delivers better performance for glitch detection and removal in SEIS very broadband records with much higher computational efficiency than existing methods.