A Deep Semantic Alignment Network for the Cross-Modal Image-Text Retrieval in Remote Sensing

Because of the rapid growth multimodal data from internet and social media, a cross-modal retrieval has become an important valuable task in recent years.The purpose is to obtain result one modality (e.g., image), which semantically similar query another text).In field remote sensing, despite great number existing works on image retrieval, there only been small amount research image-text due scarcity datasets complicated characteristics sensing data. In this article, we introduce novel network establish direct relationship between images their paired text Specifically, our framework, designed semantic alignment module fully explore latent correspondence text, used attention gate mechanisms filter optimize features so that more discriminative feature representations can be obtained. Experimental results four benchmark datasets, including UCMerced-LandUse-Captions, Sydney-Captions, RSICD, NWPU-RESISC45-Captions, well showed proposed method outperformed other baselines achieved state-of-the-art performance tasks.