Model Architectures

• Implemented CNN-based models for hierarchical spatial feature extraction.
• Developed Vision Transformer (ViT) models for capturing global dependencies across satellite images.
• Compared CNN and ViT performance to optimize accuracy and inference efficiency.

Data Processing & Augmentation

• Preprocessed satellite imagery using OpenCV and NumPy for resizing, normalization, and channel adjustments.
• Applied data augmentation techniques including rotations, flips, and color transformations to improve model generalization.

Training & Evaluation

• Trained models using TensorFlow/Keras and PyTorch frameworks.
• Evaluated using standard metrics such as accuracy, precision, recall, and F1-score.
• Visualized training and evaluation results with Matplotlib for analysis of learning curves and confusion matrices.

Pipeline Integration

• Modular design enabling end-to-end workflow from raw satellite imagery to classified land cover maps.
• Prepared for scalability to handle high-resolution satellite datasets efficiently.