Land Cover Classification Challenge

As part of our course project (Spring 2023), we embarked on a journey to tackle the challenges of land cover classification using advanced machine learning techniques.

Project Overview

This project focuses on the application of various deep learning models to the task of land cover classification. Our journey involved experimenting with different neural network architectures and domain adaptation techniques to improve the accuracy of land cover classification.

Models and Techniques Explored:

• ResNet50
• ResNeXt
• Vision Transformer
• Domain Adaptation Neural Network (DANN)

Key Milestones:

1. ResNet: Improved accuracy to 58.5% with data augmentation.
2. ResNeXt: Enhanced generalization, reaching an accuracy of 61.7%.
3. ViT: Achieved the highest result of 62.6% by experimenting with epochs.
4. Domain Adaptation (DA): Showed promising results with ResNet18, achieving 57.4% accuracy.

Note: The data was tested on OOD data, thus resulting in lower accuracy scores as typically observed

For more details, please view our final presentatation of this project.