Magnetic navigation consists in taking measure-ments of the magnetic field of the Earth which, thanks to a map of magnetic anomalies, allows us to know a precise position. However, these measurements are noisy due to the carrier making the measurements, such as an aircraft. To compensate for the magnetic disturbance of the carrier, it is possible to use the Tolles-Lawson method. However, this method does not work very well when the magnetometers are disturbed by the carrier itself, which is very often the case. To reach an accuracy of the order of ten meters, a better method is needed to manage these effects created by the carrier. Since these effects are mainly nonlinear, neural networks have become a method of choice for correcting the measurements. This paper proposes to use and compare different neural networks such as multi-layer perceptron, convolutional neural networks and recurrent neural networks for the calibration of airborne magnetometers. For this purpose, the dataset provided in the MIT magnetic navigation challenge is used to explore this problem. Compared to the Tolles-Lawson method, the use of neural networks can significantly improve the quality of the correction for aircraft effects on the measurements by up to 43%. All of the methods in this paper using neural networks were found to improve the Tolles-Lawson method. Visualization methods are also proposed to better understand the model and visualize the features that affect its prediction.
Neural Network Calibration of Airborne Magnetometers
2023-06-19
1178612 byte
Conference paper
Electronic Resource
English
Calibration of Strapdown Magnetometers in Magnetic Field Domain
Online Contents | 2006
|NTIS | 1974
|Calibration techniques for magnetometers implementing on-board de-spinning algorithms
Online Contents | 2008
|