This paper proposes a U-Net model leveraging a pre-trained VGG-16 encoder to achieve precise land cover segmentation, a critical task for environmental monitoring and urban planning. The model incorporates semantic segmentation techniques, a class-weighted cross-entropy loss function, and the Adam optimizer to address class imbalance. Pre-processing and data augmentation strategies are employed to enhance the model's adaptability to varying environmental conditions. The model was evaluated across six land cover classes urban, agricultural, forest, water, rangeland, and barren achieving an overall Intersection over Union (IoU) score of 69.27% and a mean IoU of 59.23%. Performance metrics further demonstrated high precision (83.14%), recall (74.88%), F1-score (76.78%), and accuracy (74.88%), underscoring its effectiveness. While the model successfully mitigates class imbalance, it remains computationally intensive and reliant on high-quality data. Despite these challenges, the proposed approach shows significant promise for applications in environmental research and sustainable resource management.