Earth Scan: DL for Satellite Image Analysis

The growing availability of satellite imagery has revealed immense potential for earth observation and land cover analysis. The complexity and huge amount of satellite data, however, pose advanced techniques for accurate and efficient interpretation. This project, Earth Scan: Deep Learning for Satellite Image Analysis, investigates the application of deep learning (DL) models in classifying high-resolution satellite images, specifically land cover classification using the EuroSAT dataset. The data set contains 10 classes such as Forest, River, Residential, and Industrial areas.Our intention is to see how the performance of multiple DL models, primarily VGG16 and MobileNetV2, is measured according to accuracy, efficiency, and scalability. The models are pre-trained on ImageNet and fine-tuned on the EuroSAT dataset using techniques such as data augmentation, dropout regularization, and transfer learning. The study also uses model optimization methods such as early stopping and learning rate scheduling to improve performance.The project pipeline includes preprocessing and augmenting the satellite images, model training and validation on the basis of performance metrics like accuracy and loss, and classification output analysis. By rigorous experimentation, we intend to determine the best deep learning architecture for use in remote sensing. Our research contributes to the building of more robust and scalable systems for satellite image classification.Comparison also provides insight into the merits and demerits of different architectures and guides future research in environmental monitoring, urban planning, and disaster management using satellite images.