Super-Resolution Method for Synthetic Aperture Radar Image Based on Multi-Scale Feature Extraction

Deep learning has achieved remarkable success with the super-resolution of ordinary optical images. However, synthetic aperture radar (SAR) images have unique imaging mechanisms and features different from optical images, and are faced with problems such as low signal-to-noise ratio, limited resolution, speckle noise and sidelobe, which affect the readability and quality of images. Improving SAR image quality is an important research direction in SAR image processing, and the development of deep learning technology provides a new perspective for improving SAR image quality. Deep convolutional neural networks (CNNS) or other deep learning models are usually used for training and optimization, ignoring the multidimensional features of SAR images. Therefore, we propose a SAR image super-resolution reconstruction network based on multi-scale feature extraction. By considering the multi-dimensionality of SAR image features, the proposed algorithm achieves more accurate image reconstruction, and achieves good results in both quantitative and visual evaluation.