Published June 20, 2026 | Version v1

ACCURATE GEOPHYSICAL SIGNAL RECONSTRUCTION USING BICUBIC SPLINE INTERPOLATION AND DEEP NEURAL NETWORKS WITH TRANSFER LEARNING

Description

In this paper, we provide a mathematical framework of geophysical signal analysis based on natural bicubic spline interpolation. By training the seismic data with neural networks, and interpolating that data with respect to this model, we have better results. Measuring devices cannot easily be installed in mountainous areas, water bodies, wetlands or in densely populated regions. This causes uneven and irregular data collection. We designed a training methodology, which involves the combination of digitally simulated signals and real collected signals. We first trained a deep interpolation network with a fully synthetic collection of images formed by digital simulations to train a traditional interpolation network. The feature extraction component was then frozen and the interpolation network was applied to real data sets where it was trained on incomplete collection images. Lastly, these half-assembled images were re-entered into the trained network to get interpolated output at target locations. Our research reveals that our approach is superior to the other two interpolation networks by comparing all three and prove the transfer learning performance by a series of four ablation experiments in real testing.

Files

3(3).pdf

Files (861.1 kB)

Name Size
md5:783af3b0c899a6adf5e7635f4c4c316f
861.1 kB Preview Download