An improved neural network method for aeromagnetic compensation

Abstract

Aeromagnetic compensation plays an important role in aeromagnetic data processing and can eliminate the magnetic interference generated by aircraft. Currently, aeromagnetic compensation algorithms are mostly based on linear regression equations, but linear regression has insufficient fitting function capabilities compared to neural networks. Williams proposed a compensation model from the perspective of a neural network, but this model has the problem of overfitting. Combining these two methods, this paper proposes a neural network compensation method with strong generalization. This method is based on regression equations and uses generalized regression neural networks (GRNNs) for neural network regression. Analysis of the probability density function that constitutes the GRNN shows that proper smoothing factors and multiple training sets can improve the generalization of the GRNN compensation model, thereby weakening the effect of overfitting.