Deep UNet Segmentation for High Precision Urban and Environmental Land Cover Classification

Accurate urban and environmental Land Use/Land Cover (LULC) classification from satellite imagery is essential for sustainable urban planning, environmental monitoring, and resource management. This study presents a high-precision land cover classification framework based on deep image segmentation using a U-Net convolutional neural network architecture. The proposed system processes RGB satellite images resized to a uniform resolution of 256 × 256 pixels to ensure computational efficiency and spatial consistency. Ground truth segmentation masks are converted from RGB format into categorical class labels representing five major land cover types: Urban, Water, Forest, Agriculture, and Road. To enhance model generalization and robustness, extensive data augmentation techniques, including rotation, flipping, and scaling, are applied during training. The core segmentation model employs a four-stage encoder–decoder U-Net architecture that enables hierarchical extraction of spatial and textural features through successive convolution and pooling operations, followed by symmetric upsampling and skip connections for precise boundary localization. This design effectively captures both global contextual information and fine-grained structural details in complex urban and environmental scenes. Experimental results demonstrate that the proposed Deep U-Net–based segmentation approach achieves high classification accuracy and clear class separability across diverse land cover categories. The framework provides a reliable and scalable solution for automated urban and environmental LULC mapping, supporting data-driven decision-making in geospatial analysis, smart city development, and environmental conservation applications.

Keywords: Land Use/Land Cover (LULC) Classification, Satellite Image Segmentation, Deep Learning, U-Net Architecture, Urban and Environmental Mapping