This project focuses on classifying land cover types from satellite imagery using deep learning and transfer learning techniques.
The model is trained on the EuroSAT dataset (Sentinel-2 RGB images) to identify classes such as forests, urban areas, water bodies, agricultural land, and more. Pre-trained CNN architectures like VGG16, ResNet50, and DenseNet121 are used as feature extractors and compared against a custom CNN designed for this task.
The work includes image preprocessing, model training, evaluation, and visualization of results to study performance and generalization across land-cover classes.
- Land cover classification using satellite images
- Transfer learning with pre-trained CNNs
- Custom CNN architecture for comparison
- Preprocessing and visualization of satellite imagery
- Evaluation using accuracy and class-wise performance
The project uses the EuroSAT dataset (not included in this repository).
Dataset source: https://github.com/phelber/eurosat