This project focuses on classifying geospatial satellite images using deep learning models. We compare multiple architectures and achieve a 98% accuracy with ResNet50V2.
- Data preprocessing and augmentation
- Implementation of VGG, Inception ResNet, Xception, and ResNet50V2
- Model evaluation using classification reports and confusion matrices
- Prediction on test data with probability scores
- Satellite image dataset used for classification
- Preprocessed and split into training and validation sets
| Model | Validation Accuracy | Validation Loss |
|---|---|---|
| VGG | 91% | 0.24 |
| Inception ResNet | 57% | 1.1 |
| Xception | 83% | 0.52 |
| ResNet50V2 | 98% | 0.08 |
- Classification performance is analyzed using confusion matrices.
- Model accuracy and loss are plotted for comparison.