Land Use Land Cover Classification

I have implemented three fine-tuned models ResNet-18, ResNet-50, and ViT-Base-Patch16-224 trained on the EuroSAT dataset, which contains satellite images from Sentinel-2 representing various land use and land cover classes. The models are evaluated based on their performance in classification tasks across a set of 27,000 labeled images. The dataset is split into training, validation, and test sets, with consistent label distribution maintained.

Among the models, ViT-Base-Patch16-224 performs the best, achieving the highest accuracy, precision, recall, and F1-score. It is the top performer for this dataset. ResNet-50 also delivers strong results and serves as a solid alternative for those preferring a ResNet architecture or dealing with computational constraints. Although ResNet-18 is slightly less accurate, it remains a viable option for scenarios requiring a lighter, more computationally efficient model.

This work offers a comprehensive overview of model performance, including training loss, accuracy, and evaluation metrics, along with detailed implementation instructions and dependencies.