Machine Learning-Enhanced Interpolation of Gravity-Assisted Magnetic Data

• Published in: IEEE geoscience and remote sensing letters Vol. 21; pp. 1 - 5
• Main Authors: Xu, Hong, Zhao, Lvshen, Jing, Peiqi, Yan, Jie, Zhu, Xuming, Jia, Zhuo
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Data analysis; Data interpolation; Data models; Data processing; Decoding; Deep learning; Gravity; Gravity data; Information processing; Interpolation; Interpolation methods; Machine learning; Magnetic anomalies; Magnetic data; Magnetic field measurement; Magnetic resonance imaging; Mathematical models; Observational learning; Spatial distribution
• ISSN: 1545-598X; 1558-0571
• DOI: 10.1109/LGRS.2024.3382049

The acquisition of magnetic anomaly data is generally considered a process of information degradation, with its content significantly impacting subsequent tasks involving data processing, inversion, and interpretation. Traditional interpolation methods often rely on the spatial distribution and sampling density of data, thus struggling to handle complex nonlinear relationships effectively. To address these challenges, this study employs deep learning algorithms for interpolating magnetic anomaly data, aiming to enhance the resolution of magnetic data. Additionally, gravity data are incorporated as supplementary information to improve the quality of magnetic anomaly data interpolation. Similar to magnetic data, gravity data also exhibit a certain degree of spatial correlation, as a single geological source may produce anomalies in both gravity and magnetic responses simultaneously. Through the training and prediction of deep learning networks, it is observed that the intelligent interpolation retains the subtle features of magnetic anomaly data in space while avoiding staircase-like erroneous anomalies generated by linear interpolation. Furthermore, gravity data assist in constraining the results of magnetic anomaly interpolation, enhancing their accuracy. Finally, the trained network is applied to measured data, with the input data being downsampled. The results show that the network can accurately predict magnetic anomaly data and bring them closer to the magnetic anomaly data before downsampling.