Semantic Segmentation is a fundamental discipline of computer vision that aims to divide an image into distinct segments and provide semantic labels for every pixel. Semantic segmentation models consist of deep convolutional neural networks that continuously show striking improvements in the techniques of image analysis. This study applied different deep learning models to high-resolution satellite imagery of the Dubai metropolitan area. 72 images were utilized for this study, which were collected from the Pleiades-1A satellite. After augmentation, the total images for analysis were 504. The images considered in this study were carefully annotated and pre-processed to improve the training and reliability of the models. Five distinct deep learning models, VGG16, DeepLabV3+, Inception-ResNet-V2, Multi-UNet, and MobileNet-V2, were applied and compared to find the best-performed model among them. Inception-ResNet-V2 outperformed other models with 92% accuracy and a dice coefficient of 0.87, proving its superiority in extracting intricate urban features. The predicted mask image produced by the Inception-Resnet-V2 showed better results compared to other models’ predicted masks. This study emphasised the importance of identifying the most suitable structure and training regimen that significantly extracts the information from an image. Future studies should focus on improving the model structure to increase the model accuracy and reduce the execution time substantially.