Advancing Satellite Image Segmentation: A Deep Learning Approach with U-Net, FPN, and DeepLabV3+

This research paper explores the automated categorization and segmentation of land cover from satellite images using deep learning models, mainly U-Net with ResNet50, FPN, and DeepLabV3+. The challenge utilizes the Albumentations library for ground truth augmentation and uses metrics such as Intersection over Union (IoU) and Dice Coefficient to evaluate segmentation performance. The study applies the models to classify various land types inclusive of urban, agriculture and woodland land whilst specializing in sustainable development, city planning and agriculture. A ground fact mask is hired for schooling, while the Adam optimizer and cross-entropy loss guide the model's learning technique. Results exhibit the ability of segmentation models in appropriately figuring out land cover sorts, imparting a sensible tool for environmental and making plans groups.