A Hierarchical Approach to Remote Sensing Scene Classification

Remote sensing scene classification deals with the problem of classifying land use/cover a region from images. To predict development and socioeconomic structures cities, status use in regions is tracked by national mapping agencies countries. Many these land-use types that are arranged multiple levels. In this paper, we examined efficiency hierarchically designed convolutional neural network (CNN)-based framework suitable for such arrangements. We NWPU-RESISC45 dataset our experiments arrange data set two-level nested hierarchy. Each node hierarchy trained using DenseNet-121 architectures. provide detailed empirical analysis to compare performances hierarchical scheme its non-hierarchical counterpart, together individual model performances. also evaluated performance structure statistically validate presented results. The results show although classifiers different sub-categories perform considerably well, accumulation errors cascaded prevents exceeding deep model.