Automated Land Cover Classification in Urban Environments with Deep Learning-Based Semantic Segmentation

Partisan symbolism has become a vital asset for various applications, including land cover classification, urban planning, disaster management, and environmental monitoring. In this study, a deep learning approach is proposed for the semantic segmentation of satellite images. The proposed model leverages a multi-level U-Net architecture, which has shown effectiveness in various image segmentation tasks. The dataset used in this study comprises satellite images with corresponding ground truth masks labeling different land cover classes, such as buildings, roads, vegetation, water bodies, and unlabeled regions. The data is preprocessed, and the ground truth masks are encoded into categorical labels to facilitate model training. The model is trained using a combination of pixel-wise crossentropy loss and the Jaccard coefficient as evaluation metrics to optimize performance. Test results demonstrate the effectiveness of the proposed approach in accurately segmenting satellite images, achieving high accuracy and Jaccard coefficient values on both training and validation datasets. Additionally, a comprehensive analysis of the model’s performance is provided, discussing potential applications and future directions for satellite image segmentation tasks. The trained model is saved for further deployment and integration into satellite image processing pipelines, offering valuable insights for various remote sensing applications.