Dual-Convolutional Enhanced Residual Network for Single Super-Resolution of Remote Sensing Images

The image super-resolution aims to recover a high-resolution image using a single or sequential low-resolution images. The super resolution methods based on deep learning, especially the deep convolutional neural network, have achieved good results. In this paper,we propose Dual-Convolutional Enhanced Residual Network (DCER) for remote sensing images based on residual learning, which concatenates the feature maps of different convolutional kernel sizes (3 \(\times \) 3, 5 \(\times \) 5). On the one hand, it can learn more high-frequency detail information by combining the local details of different scales; on the other hand, it reduces network parameters and greatly shorten the training time. The experimental results show that DCER achieves favorable performance of accuracy and visual performance against the state-of-the-art methods with the scale factor 2x, 4x and 8x.