Automated Processing of a Large-Scale Airborne Electromagnetic Survey by Deep Learning

Publikation: KonferencebidragPaperForskningpeer review

Abstract

Modern airborne electromagnetic (AEM) surveys generate large data sets spanning over thousands of line kilometres. Some portions of these data sets are often unusable due to couplings with infrastructure, such as power lines and fences. The inversion of coupled data results in spurious subsurface features, which may give erroneous geological interpretations and, in turn, biased decisionmaking. To mitigate this, the corrupted data are often culled manually before inversion, which is a labour-intensive and time-consuming task, and requires specialized expertise. To address this challenge, we have developed a deep learning expert system that automates the AEM data processing workflows. Our method employs a deep convolutional auto-encoder to identify corrupted data and is designed to generalize across diverse geological conditions and survey areas. In this study, we evaluate the generalization performance of our method on a large AEM survey conducted in Northland, New Zealand. Our approach processes 6471 line kilometres of data in 15 minutes and identifies corrupted data, showing strong spatial correlation. The inversion results reveal very few potential anomalies, which are undergoing manual inspection. Overall, our proposed approach demonstrates high-quality data processing with minimal quality inspection required, making it a promising solution to automate AEM data processing workflows.

OriginalsprogEngelsk
Publikationsdato2023
Antal sider5
DOI
StatusUdgivet - 2023
BegivenhedNSG2023 29th European Meeting of Environmental and Engineering Geophysics - Edinburgh , Storbritannien
Varighed: 3 sep. 20237 sep. 2023
Konferencens nummer: 29

Konference

KonferenceNSG2023 29th European Meeting of Environmental and Engineering Geophysics
Nummer29
Land/OmrådeStorbritannien
ByEdinburgh
Periode03/09/202307/09/2023

Fingeraftryk

Dyk ned i forskningsemnerne om 'Automated Processing of a Large-Scale Airborne Electromagnetic Survey by Deep Learning'. Sammen danner de et unikt fingeraftryk.

Citationsformater