Remote Sensing Cross-Modal Text-Image Retrieval Based on Global and Local Information

Cross-modal remote sensing text-image retrieval (RSCTIR) has recently become an urgent research hotspot due to its ability of enabling fast and flexible information extraction on (RS) images. However, current RSCTIR methods mainly focus global features RS images, which leads the neglect local that reflect target relationships saliency. In this article, we first propose a novel framework based (GaLR), design multi-level dynamic fusion (MIDF) module efficaciously integrate different levels. MIDF leverages correct information, utilizes supplement uses addition two generate prominent visual representation. To alleviate pressure redundant targets graph convolution network (GCN) improve model s attention salient instances during modeling features, de-noised representation matrix enhanced adjacency (DREA) are devised assist GCN in producing superior representations. DREA not only filters out with high similarity, but also obtains more powerful by enhancing objects. Finally, make full use similarity inference, come up plug-and-play multivariate rerank (MR) algorithm. The algorithm k nearest neighbors results perform reverse search, improves performance combining multiple components bidirectional retrieval. Extensive experiments public datasets strongly demonstrate state-of-the-art GaLR task. code method, MR algorithm, corresponding files have been made available at https://github.com/xiaoyuan1996/GaLR .