Geomagnetic reference map super-resolution using convolutional neural network

Abstract

Geomagnetic matching navigation technology, as an important assisted navigation method, has gradually been wildly used in the navigation field, in which the quality of geomagnetic reference map is of great significance on improving the guidance and positioning accuracy. However, the existing geomagnetic reference map reconstruction methods can hardly satisfy the actual needs. Considering the popularity of deep learning methods in the image super-resolution (SR) field, a geomagnetic reference map SR method using a convolution neural network is proposed. First, rectangular harmonic analysis theory is used to construct the training dataset, which is based on data from the United States Geological Survey and the International Geomagnetic Reference Field. Then, a multi-channel convolutional neural network, which integrates all the three independent components in the geomagnetic reference field, is proposed of reference map SR reconstruction. Experimental results show the effectiveness of our proposed method over other state-of-the-art methods.