An innovative super-resolution method for remote sensing images is proposed,capable of supporting arbitrary magnification and effectively restoring spatial details of remote sensing images,thus offering higher flexibility for diverse applications.The method is based on the spectral-scale attention mechanism,a single neural network is utilized to process images at arbitrary magnifications.Network weights and depth-wise separable meta up-sample are generated dynamically,high-precision image reconstruction is realized while network parameters and computational complexity are significantly reduced.Extensive experimentation on four remote sensing image datasets indicates the approach per-forms better in both quantitative metrics and visualization quality.
关键词
遥感图像/任意倍率超分辨率/深度学习/光谱尺度注意力/深度分离元上采样
Key words
remote sensing images/super-resolution at arbitrary magnification/deep learning/spectral-scale attention/depth-wise separable meta up-sample
维普
万方数据