DFEN: Dual Feature Enhancement Network for Remote Sensing Image Caption

The remote sensing image caption can acquire ground objects and the semantic relationships between different objects. Existing algorithms do not enough object information from remote-sensing images, resulting in inaccurate captions. As a result, this paper proposes codec-based Dual Feature Enhancement Network (“DFEN”) to enhance both text levels. We build Image-Enhancement module at level using multiscale characteristics of images. Furthermore, more discriminative context features are obtained through module. hierarchical attention mechanism aggregates multi-level supplements ignored due large-scale differences. At level, we use image’s potential visual guide Text-Enhance module, guidance that correctly focus on Experiment results show DFEN model images text. Specifically, BLEU-1 index increased by 8.6% UCM-caption, 2.3% Sydney-caption, 5.1% RSICD. has promoted exploration advanced semantics facilitated development caption.