AI on AI: Training AI classifier on AI-generated remote sensing images

Over the past few years, the rapid technological development in the field of imaging and aerospace technology has driven advancements in the remote sensing field. This has opened lots of new opportunities. However, along with these continuous advancements in this field, some challenges arise. One of these challenges is the limited accessibility to high accuracy remote sensing datasets, that are meant to train AI classifier models. Thus, we propose to use image generating AI models to generate dummy remote sensing images having real-life images' features, to substitute the shortage of remote sensing images. In this project, Dall-E 2 was used to generate the images of seaports and football stadiums. Which were then used to train an AI classifier. Although the model trained on generated images only, it classified 90.2% of real remote sensing images correctly, proving its practicality; however, further efforts can be invested to improve the reliability of this model. This project will open the chance for students and researchers to get easy access to remote sensing images that can be harder to access in normal conditions.