SE-HRNet: A Deep High-Resolution Network with Attention for Remote Sensing Scene Classification

Remote sensing scene classification is a challenging task due to the complex high-resolution textures and similarity within and variations between classes. The high-resolution network (HRN et) is introduced into this application considering its capability of high-resolution and multi-scale semantic representations. Moreover, squeeze-and-excitation (SE) attention mechanism is employed to enhance and suppress features based on their interdependence between channels. By fusing SE block into HRNet, the proposed SE-HRNet can better distinguish scene categories with enriched features. Experimental results on the NWPU-RESIS45 and AID data sets validate the state-of-the-art performance of the proposed SE-HRNet.