VLCA: vision-language aligning model with cross-modal attention for bilingual remote sensing image captioning

In the field of satellite imagery, remote sensing image captioning (RSIC) is a hot topic with challenge overfitting and difficulty text alignment. To address these issues, this paper proposes vision-language aligning paradigm for RSIC to jointly represent vision language. First, new dataset DIOR-Captions built augmenting object detection in optical (DIOR) images manually annotated Chinese English contents. Second, Vision-Language model Cross-modal Attention (VLCA) presented generate accurate abundant bilingual descriptions images. Third, cross-modal learning network introduced problem visual-lingual Notably, VLCA also applied end-to-end captions generation by using pre-training language Chinese. The experiments are carried out various baselines validate on proposed dataset. results demonstrate that algorithm more descriptive informative than existing algorithms producing captions.