Performance Evaluation Of Convolutional Neural Network And Generative Adversarial Network For Super Resolution Of Satellite Image

Abstract

As a possible reaction for the issues related with low-resolution imagery in remote sensing applications, deep learning-driven super-resolution (SR) approaches have become unmistakable. The bound spatial resolution of satellite images, which are essential for calamity the board, metropolitan new turn of events, and typical monitoring, is by and large refined through air and sensor blocks. While methods bilinear and bicubic contribution have been used historically to besides support image resolution, they decidedly scorn staying aware of little nuances and part reliability. This paper isolates these customary systems and the sensibility of two deep learning models: Convolutional Neural Network (CNN) and Generative Adversarial Network (GAN). We evaluate enhancements to the extent that Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) by scaling low-resolution images through the preparation of deep learning models on colossal satellite image datasets, expressly acquired from repositories like Landsat, Sentinel. Then again, with normal strategies like bilinear and bicubic inclusion, the suggested models yield an ordinary PSNR improvement of 15-20% and a SSIM move of 10-12%. Further made highlight extraction, crisper photos, and more exactness in typical assessment and land cover gathering are the deferred outcomes of these updates. In outline, the SR models driven by deep learning show brilliant commitment in regards to disturbing satellite imaging and giving a way forward to geospatial data dealing with that is both more careful and reasonable.