Building Detection from Satellite Images Using Deep Learning

Abstract

Satellite photography has transformed our capacity to comprehend and address dynamic alterations in our surroundings. Automated identification of buildings in satellite imagery is essential for urban planning and disaster management purposes. Conventional techniques frequently face challenges when dealing with the intricate and diverse nature of satellite imagery, requiring the use of more sophisticated methods. The research aims to improve building detection accuracy by utilizing deep learning methods, primarily Convolutional Neural Networks (CNNs, and more specifically the YOLOv8 architecture). The suggested methodology includes creating and using a CNN model, YOLOv8, designed to properly detect buildings and evaluate building damage levels in satellite pictures. The YOLOv8 architecture is intricately crafted to capture complex spatial hierarchies found in satellite data. Training and validation processes are carried out utilizing annotated datasets, with iterative adjustments made to enhance model performance. Postdisaster analysis is a crucial stage where a CNN model created using YOLOv8 is used to assess building damage and comprehend the effects of natural catastrophes on urban infrastructure. Heatmaps are created as visual aids to show the level of harm, helping stakeholders make decisions. Continual enhancements to the model are guided by analysis conducted after a disaster, ensuring continuous development and flexibility. The project intends to enhance building detection approaches using deep learning, particularly YOLOv8 based Convolutional Neural Networks, for disaster response and urban planning. Combining CNNs with satellite images provides a potent tool for spatial analysis and decision-making, impacting emergency management and sustainable urban development.