A Denoising Method Based on Kernel Principal Component Analysis for Airborne Time‐Domain Electromagnetic Data

Abstract

Airborne time-domain electromagnetic (ATEM) data usually contain natural and cultural noise, which can lower data quality, influence inversion accuracy or even lead to incorrect interpretation if it is not removed from data using an appropriate filter. To solve this problem, this work suggests a denosing method based on kernel principal component analysis. Firstly, it extracts the principal component from stacked decay curves. Then the useful signals, which are associated with subsurface media, and noise are separated using the energy ratio. Finally, these signals are used to perform reconstruction. This method can not only remove natural noise such as spikes or oscillation caused by sferies, but also effectively suppress cultural noise. Using this method and the AeroTEM software, the real ATEM data from a helicopter survey is processed separately. Comparison of the results shows that the denoising effect of the method suggested by this paper is superior to that of the AeroTEM software.