The super-resolution reconstruction of SAR image based on the improved FSRCNN

Abstract

Synthetic aperture radar (SAR) images have become an important way to obtain information in the military and civilian fields, because of its unique advantages. With the development of technology and the need of application, people subjectively put forward a higher demand for image quality. Image resolution is a key factor for evaluating digital image quality and is the basis for subsequent image processing. However, the image quality of SAR images is far worse than that of optical images because of the imaging mechanism and so on. Therefore, it is more difficult to realise super-resolution reconstruction on SAR images. In the image super-resolution reconstruction method, a reconstruction-based method is generally used, but the effect is poor. A method based on deep learning is used to realise the reconstruction of SAR images based on floating-point data by obtaining the mapping relationship between low-resolution images and high-resolution images. At the same time, the SSIM index is introduced into the loss function, so that the reconstructed SAR image is improved both in subjective visual and in objective evaluation indicators. It lays the foundation for subsequent SAR image recognition and other work.

Keywords

• optical images
• synthetic aperture radar
• image resolution
• image recognition
• image reconstruction
• image processing
• radar imaging
• image resolution
• digital image quality
• subsequent image processing
• optical images
• imaging mechanism
• image super-resolution reconstruction method
• reconstruction-based method
• low-resolution images
• high-resolution images
• reconstructed SAR image
• subsequent SAR image recognition
• synthetic aperture radar images