A New Method of Wave Parameter Retrieve Based on Transformer

A new method using deep learning to retrieve the main parameters of ocean waves which include significant wave heights, peak wave periods, peak wave lengths and peak wave directions. At present, retrieve of significant wave height can use the model of Convolutional Long Short-Term Memory(ConvLSTM) with attention. However, the accuracy of this method needs to be improved, and it can only retrieve a single wave parameter. We propose a Transformer-based network structure, which improves the accuracy effectively and can simultaneously retrieve multiple wave parameters. Finally, we use Pearson's correlation coefficient as an evaluation index to compare with the baseline. Experimental results show that the performance of the network has been improved.