The rise of satellite imagery data in the modern world has opened new opportunities in modern challenges such as disaster relief, urban planning, and environmental monitoring. Semantic segmentation is one of the crucial methods in deep learning to extract comprehensive land cover data. Using the DeepGlobe LandCover Classification dataset, this research applies multiple deep learning models on the task of semantic segmentation using U-NET, Fast-SCNN, and DeepLabV3+. We came up with an entirely new approach by dividing images into patches substantially minimizing computational intensity yet offering segmentation accuracy in excess of that in the unpatched scenarios. This study proposes an optimized U-NET-based scaled-down version utilizing this alongside improving the deployability of a model on resource-constrained systems, such as an edge device and drones. The performance of this lightweight adaptation is contextualized using DeepLabV3+ and Fast-SCNN, which are known for their exceptional accuracies. The research findings demonstrate the potential of the enhanced U-NET variant for edge applications by showing that they strike a balance between accuracy and efficiency.