Generative and Deep Learning Analysis for Sharpening and Detection of Infrastructure from Satellite Imagery

Satellite imagery plays an important role in urban planning and infrastructure management, making it necessary for accurate analysis and decision-making utilizing high-quality visuals. Building upon our previous study on automated satellite imagery generation using Diffusion models, this paper focuses on enhancing the clarity and hence the practical usability of these generated images through GFPGAN-based sharpening. Furthermore, we also explore the application of rooftop detection using three methodologies: direct YOLO processing, achieving 73% accuracy, the Slicing Aided Hyper Inference (SAHI) methodology for YOLO, achieving 68 % accuracy, and a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$3 \times 3$</tex> grid-based image segmentation approach, achieving an overall 70 % accuracy. A comparison of these methods is presented, highlighting their metrics such as accuracy and precision. This study aims to compare the feasibility of image enhancement and detection pipelines using AI models for real-world urban development applications.