Automated Land Use classification from satellite imageryA Deep learning

This project introduces an automated land use and land cover classification system based on high-resolution satellite imagery and deep learning. Trained on the DeepGlobe Land Cover Classification Dataset, which consists of paired RGB satellite images and annotated masks, a U-Net++ semantic segmentation model is utilized to classify land into seven categories: urban land (man-made built-up areas), agriculture land (farms and plantations), rangeland (open green areas), forest land (tree-covered regions), water (rivers, lakes, and wetlands), barren land (unvegetated rocky or sandy areas), and unknown. The U-Net++ architecture, featuring nested skip connections, improves segmentation accuracy by effectively capturing multi-scale features in complex landscapes. To make the system accessible, a web application is developed with user registration and login functionality, allowing authenticated users to upload satellite images and receive accurate predicted land cover masks. This tool supports environmental monitoring, urban planning, and sustainable land management applications. (148 words)

Keywords

Land Cover Classification, Satellite Imagery, Deep Learning, U-Net++, Semantic Segmentation, DeepGlobe Dataset, Web Application, Remote Sensing, Land Use Mapping, Convolutional Neural Networks