An atmospheric refractivity inversion method based on deep learning

Abstract An atmospheric refractivity inversion method based on deep learning is introduced. Atmospheric duct is an anomalous phenomenon of electromagnetic waves in the atmosphere that affects the use of radio equipment, obtaining real-time atmospheric duct information is of great significance for ship communication, navigation and radar detection. In order to achieve a real-time inversion system, a multilayer perceptron (MLP), a quintessential deep learning model is chosen as the inversion method. After trial-and-error, a five-hidden-layer MLP with rectified linear unit activation function is chosen. Since refractivity inversion is a regression problem, the mean-squared error is utilized to construct the loss function, and the adaptive moment estimation (Adam) algorithm is chosen to accelerate the training convergence. A pregenerated database is used to train the MLP, and thus invert the refractivity profile. The results demonstrate the feasibility and efficiency of this MLP-based inversion method.