In this paper, the performances of various deep learning models, including YOLOv8, ResNet, and VGG across a spectrum of configurations and training scenarios are studied. The aims of this study are (i) to develop and implement a deep learning model for accurately detecting building damage in satellite images and (ii) to develop methodologies for quantifying and analysing building damage level. By comparing with various Convolutional Neural Network (CNN) models, this research aims to improve the accuracy and efficiency of building detection and damage assessment, ultimately aiding in urban planning and disaster response efforts. This comparative analysis not only highlights the relative advantages of each model but also guides the selection of appropriate deep learning techniques for enhancing building detection accuracy. By optimizing these deep learning models, we aim to advance the capabilities of satellite image-based monitoring systems, thus supporting more efficient urban planning and more responsive disaster management strategies. © 2025 IEEE.