PIX2PIX GAN for Satellite and Map Image Conversion
Tejashwini R1, Dr. T Vijaya Kumar2
1 Student, Department of MCA, Bangalore Institute of Technology, Karnataka, India
2Professor, Department of MCA, Bangalore Institute of Technology, Karnataka, India
Abstract
In recent years, the demand for advanced methods of converting geographical data between different formats has increased rapidly. Urban planners, researchers, and environmental scientists often need to switch between abstract maps and highly detailed satellite imagery to carry out meaningful analysis. However, existing systems are mostly manual, costly, and not suitable for real-time applications. They also lack the intelligence to preserve fine details such as roads, rivers, vegetation boundaries, and urban structures. This research addresses these shortcomings by presenting a Pix2Pix Generative Adversarial Network (GAN)-based image translation system that is capable of performing bidirectional conversions between maps and satellite images.
The proposed system makes use of a dual generator architecture, one trained for map-to-satellite translation and another for satellite-to-map translation. With this approach, the system produces outputs that are visually realistic while maintaining spatial accuracy. Unlike traditional GIS-based approaches that demand high technical expertise and expensive licensing, this solution is accessible through a simple web-based interface built with Flask, making it practical for a wide range of users, including students and professionals who are new to GIS technologies. Preprocessing and postprocessing pipelines are integrated to enhance image quality, reduce artifacts, and ensure consistency in outputs.
This paper highlights how Pix2Pix GAN can bridge the gap between conventional cartographic methods and modern deep learning technologies. It demonstrates how the system can generate reliable results in under ten seconds per image, making it suitable for near real-time applications. The model has potential applications in areas such as urban development, disaster management, environmental monitoring, and education. While the results are promising, the study also acknowledges challenges related to data availability, scalability, and generalization across unseen regions.
Key words: Pix2Pix Generative Adversarial Network (GAN), Image-to-Image Translation, Map-to-Satellite Conversion, Satellite-to-Map Conversion, Deep Learning, Geographical Information Systems (GIS), Remote Sensing, Urban Planning, Cartographic Visualization.
