Speaker Bio & Abstract
BiographyMSc Geospatial Technologies Student (2021-2023). Bachelors in Computer Science. 4 Years of Experience in machine learning and developing spatial products. Awarded best product in Geomundus Conference 2021AbstractSuper Resolution for Satellite ImageryIn recent years, deep learning has quickly evolved to be the go-to solution for any kind of analysis on non-linear data. One such usage has been that of Generative Adversarial Networks (GAN) in the field of Computer Vision. GAN models have a variety of applications for image processing, specifically, super-resolution of images. A lot of work has been done to enhance or upscale generic RGB imagery such as the ones taken from a mobile or digital camera. However, in the field of remote sensing, it presents challenges like preserving the resolution of the original sensor and tuning the algorithms to expect values beyond 255. We cannot simply change the size of a satellite image using a common interpolation methodology. From data preparation to enhancing a complete set of tiles at scale, the upsampling/downsampling requires the ratio of number pixels to an sq km area to be preserved. This tool, developed using an open-source algorithm of SRGAN was tuned to work for satellite imagery, resulting in upscaling the resolution of any sensor by 4 times. In this example, two sensors, SPOT, and Sentinel-2 were enhanced to 0.6m and 2.5m respectively.