Text Semantic Fusion Relation Graph Reasoning for Few-Shot Object Detection on Remote Sensing Images

Most object detection methods based on remote sensing images are generally dependent a large amount of high-quality labeled training data. However, due to the slow acquisition cycle and difficulty in labeling, many types data samples scarce. This makes few-shot an urgent necessary research problem. In this paper, we introduce method text semantic fusion relation graph reasoning (TSF-RGR), which learns various relationships from common sense knowledge end-to-end manner, thereby empowering detector reason over all classes. Specifically, region proposals provided by basic network, first build corpus containing number language descriptions, such as attributes relations, used encode corresponding embeddings for each region. Then, structures constructed between regions propagate learn key spatial relationships. Finally, joint module is proposed actively enhance reliability robustness feature representation focusing degree influence different relations. Our TSF-RGR lightweight easy expand, it can incorporate any form information. Sufficient experiments show that information introduced deliver excellent performance gains baseline model. Compared with other detectors, achieves state-of-the-art shot settings obtains highly competitive results two benchmark datasets (NWPU VHR-10 DIOR).