An Intra-Class Ranking Metric for Remote Sensing Image Retrieval

With the rapid development of internet technology in recent years, available remote sensing image data have also been growing rapidly, which has led to an increased demand for retrieval. Remote images contain rich visual and semantic features, high variability complexity. Therefore, retrieval needs fully utilize information perform feature extraction matching. Metric learning widely used as it can train embedding spaces with discriminability. However, existing deep metric methods learn discriminability by maximizing differences between classes, while ignoring inherent intra-class during process. In this paper, we design a new sample generation mechanism generate samples from positive that meet boundary constraints, thus obtaining quantifiable real samples. Based on relationship, use self-supervised approach ranking loss function, improves generated space same class maintains their relationship space. Moreover, function be easily combined methods. Our aim is help network better extract features further improve performance through loss. Finally, conduct extensive experiments multiple remote-sensing datasets using evaluation metrics such mAP@K, demonstrate sample-generated effectively