Unveiling the Landscape: A Comparative Study of U-Net Models for Geographical Features Segmentation

Geographical features segmentation is a critical task in remote sensing and earth observation applications, enabling the extraction of valuable information from satellite imagery and aiding in environmental analysis, urban planning, and disaster management. The U-Net model, a deep learning architecture, has proven its efficacy in image segmentation tasks, including geographical feature analysis. In this research paper, a comparative study of various U-Net models customized explicitly for geographical features segmentation is presented. The study aimed to evaluate the performance of these U-Net variants under diverse geographical contexts and datasets. Their strengths and limitations were assessed, considering factors such as accuracy, robustness, and generalization capabilities. The efficacy of integrated components, such as skip connections, attention mechanisms, and multi-scale features, in enhancing the models’ performance was analyzed.