A Gramian Angular Field and Dual-Stream Transformer-Based Inversion Model for Retrieving Evaporation Duct from Radar Sea Clutter

This article builds a novel evaporation duct inversion model based on the Gramian angular field (GAF) and dual-stream Transformer, termed EVDIFormer. Considering the complex nonlinear correlations between sea clutter and modified refractivity profile (M-profile) within the evaporation duct, this model leverages Transformer's superior feature extraction abilities to model the mapping between them. Moreover, the model will transform input sea clutter into two different sea clutter GAF matrices before feature extraction. This approach makes the model more efficient in processing sea clutter data by transforming complex nonlinear features into sparse matrix representations that can be handled more easily. By adopting dual-stream Transformer to extract features separately from the different GAF matrices, the model can extract and fuse different sea clutter features, thus achieving a more comprehensive feature extraction of sea clutter.