SGMNet: Scene Graph Matching Network for Few-Shot Remote Sensing Scene Classification

Few-Shot Remote Sensing Scene Classification (FSRSSC) is an important task, which aims to recognize novel scene classes with few examples. Recently, several studies attempt address the FSRSSC problem by following few-shot natural image classification methods. These existing methods have made promising progress and achieved superior performance. However, they all overlook two unique characteristics of remote sensing images: (i) object co-occurrence that multiple objects tend appear together in a (ii) spatial correlation these are distributed some structure patterns. Such very beneficial for FSRSSC, can effectively alleviate scarcity issue labeled images since provide more refined descriptions each class. To fully exploit characteristics, we propose graph matching-based meta-learning framework called SGMNet. In this framework, construction module carefully designed represent test or class as graph, where nodes reflect meanwhile edges capture correlations between objects. Then, matching further developed evaluate similarity score Finally, based on scores, perform prediction via nearest neighbor classifier. We conduct extensive experiments UCMerced LandUse, WHU19, AID, NWPU-RESISC45 datasets. The experimental results show our method obtains performance over previous state-of-the-art