Prediction of sea clutter characteristics by deep neural networks using marine environmental factors

The fundamental characteristics of sea clutter play a crucial role in clutter suppression and target detection for marine radar. Besides the radar parameter, the characteristics of sea clutter are determined by the marine environmental factors essentially. As a supplement to the high-cost field experimental measurement and time-consuming physical modeling based on electromagnetic scattering theory, several attempts have been made to predict sea clutter amplitude by artificial neural network (ANN) models using the radar measured data. Compared with the directly measured sea clutter data, marine environmental factors such as wind speed, significant height of wave etc., can be obtained more easily and efficiently. This study set out to examine the feasibility of predicting the characteristics of sea clutter directly by the marine environmental factors. Deep neural network (DNN) models to predict individually mean amplitude, Doppler shift, and Doppler broadening of sea clutter are proposed in this work, which take wind speed, significant height of wave and mean period of wave as input. The training and test set is generated by a hybrid scattering model. The experimental results indicate that the prediction accuracy of HH-polarized mean amplitude, Doppler shift, and Doppler broadening are 97.8, 98.9 and 89.0%, respectively, and the corresponding accuracy of VV-polarized are 94.8, 97.8 and 89.5%. These findings suggest the proposed data-driven model using marine environmental factors directly is a competitive candidate in predicting sea clutter characteristics; therefore, it can be applied to complement and extend the existing measured and calculated data sets.