Semantic Segmentation for Land Cover Using Bhuvan 2D Satellite Images

This research study presents a a novel approach for semantic segmentation of land cover using Bhuvan 2D satellite images. The proposed approach utilizes U-Net models for classification. The dataset consists of 2D satellite images covering a region in India. A pre-processing pipeline is applied to extract relevant features and generate ground-truth labels. The model is trained on the dataset, and the results demonstrate its effectiveness in accurately classifying different types of land cover, such as forest, water bodies, agriculture, roads and urban areas, using performance metrics such as Dice coefficient (F1-Score), and Intersection over Union (IoU). The model is made more accurate and robust through the implementation of data augmentation techniques. The proposed approach can have significant applications in environmental monitoring, disaster management, and urban planning, providing accurate and detailed information on land cover patterns and changes over time.