Edge-enhanced efficient network for remote sensing image super-resolution

• DOI: 10.1080/01431161.2022.2128924
• Publication Date: 2022-10-19T13:35:39Z
• Authors: Tianlin Zhang; Hongzhen Chen; Shi Chen; Chunjiang Bian
• Keywords: Convolution (computer science); Edge device; Boosting

The super-resolution (SR) reconstruction is drawing increasing attention in remote-sensing image processing, owing to improving the spatial resolution and enriching details of initially obtaining low-resolution (LR) images. Different from natural images, remote sensing images usually have more complex distribution degradation processes. Although current deep convolutional neural network (DCNN)-based approaches reach better performance by deepening introducing mechanism, this accompanied parameters, computation, designs. To improve efficiency SR we proposed edge-enhanced efficient (EESR). Specifically, design an inception-like multi-branch convolution block, named block (EEB), enrich edge-aware capabilities network, which includes multi-order gradient extraction other feature enhancement branches. Furthermore, re-parameterization introduced into inference stage aiming at boosting inference. We re-parameterize weights EEB (EEC) via equivalent transformation reduce parameters computational cost model. In addition, construct a paired dataset GF2-Port for with simulation, based on data GaoFen-2 port strait. Extensive experiments both established public datasets indicate that EESR outperforms comparable terms balancing restoration accuracy efficiency.