一种基于点扩散函数的卫星影像超分重建方法OA

Satellite imagery super-resolution reconstruction technology is a key method for improving image resolution and clarity at low cost.To address the difficulty of obtaining training samples for deep learning-based image super-resolution tasks,and to overcome the shortcomings of existing super-resolution methods in modeling the real degradation process of sat-ellite imagery,this paper designed a satellite image-specific degradation model based on the point spread function(PSF).A two-dimensional elliptical Gaussian PSF was adopted to simulate motion blur in satellite imagery.The degradation process was fully simulated through a linear combination of blurring,downsampling,and additive noise.To address the bottleneck of scarce paired training data in remote sensing image super-resolution,a cross-domain synthetic training strategy was pro-posed:high-resolution(HR)aerial imagery was used as input to simulate the degradation process,generating low-resolution(LR)satellite images and constructing HR-LR sample pairs.The synthetic dataset was used to iteratively optimize and train the real-world enhanced super-resolution generative adversarial networks(Real-ESRGAN),achieving high-quality recon-struction of satellite imagery.