Recurrent Neural Network-Based Approach for Sparse Geomagnetic Data Interpolation and Reconstruction

Abstract

Aimed to interpolate the geomagnetic data from under-sampled or missing traces, this paper presented an approach based on recurrent neural network (RNN) techniques to avoid the time & labor-intensive nature of the traditional manual and linear interpolation approaches. In this paper, a deep learning algorithm, long short-term memory (LSTM) was employed to build the precisely model for sparse geomagnetic data interpolation. First, a continuous regression hyperplane was specified to recognize the probably intrinsic relationships between sparse and integral traces by inputting the training data. Afterward, the trained model was tested with 20% of the trained geomagnetic data and other new untrained data for validation. Finally, extensive experiments were conducted for 2D and 3D field data. The results demonstrated that our RNN-based approach was more superior than a classic linear method and a state-of-the-art method, support vector machine (SVM), as the interpolation precision was approximately improved by 10%.