At Institute of Infrastructure, Technology, Research and Management, I led a summer research project on satellite image super-resolution using ESRGAN. I prepared and preprocessed OLI-2 MSI imagery, then trained and refined the model to enhance lower-resolution images.
On my RegionAdaptiveFusion-BasedGAN project, I designed a fusion-based GAN for satellite imagery and curated a dataset using 10m ISRO imagery and 5m ground-truth targets. The model achieved a PSNR of 37.41 dB and an SSIM of 0.9477 on the custom dataset.
I also adapted ESRGAN for the OLI2MSI dataset, training it from scratch with PyTorch to achieve 4x image resolution and an average PSNR of 33.64 dB. I’m preparing a Q1/Q2 journal research paper based on my BTP project.
