EuroSAT Dataset

Overview

This dataset contains satellite images from the EuroSAT datasetThe dataset consists of RGB images with 10 different classes, each representing a distinct type of land use.

Dataset Summary

Classes: 10 (e.g., Annual Crop, Forest, Herbaceous Vegetation, Highway, Industrial, Pasture, Permanent Crop, Residential, River, Sea/Lake)
Number of Images: 27,000+ images split into training and validation sets
Image Size: 64x64 pixels, 3 channels (RGB)
Data Augmentation: Random flipping, random rotation, and normalization

Usage

This dataset is ideal for training and fine-tuning image classification models, specifically for applications in remote sensing, urban planning, environmental monitoring, and agricultural management.

Key Features

Preprocessing: Images were preprocessed using TensorFlow, including resizing, normalizing, and applying augmentation techniques to improve model robustness.
Model: Fine-tuned with Vision Transformer (ViT) to leverage attention-based mechanisms for superior performance in image classification tasks.

Dataset Structure

Training Set: Approximately 80% of the dataset
Validation Set: Approximately 20% of the dataset
Format: Images are stored as PNG files with associated labels.

Labels

Class labels are encoded as integers and correspond to the following categories:
1. Annual Crop
2. Forest
3. Herbaceous Vegetation
4. Highway
5. Industrial
6. Pasture
7. Permanent Crop
8. Residential
9. River
10. Sea/Lake

Practical Applications

Urban Planning: Identifying residential and industrial areas from satellite imagery.
Agriculture: Monitoring crop types and assessing agricultural land use.
Environmental Protection: Tracking changes in natural land covers like forests and rivers.

How to Use

To use this dataset, you can directly load it through the Hugging Face datasets library:

from datasets import load_dataset

dataset = load_dataset("MuafiraThasni/eurosat-dataset-with-image")

Reference

EuroSAT dataset is originally proposed in:

@misc{helber2019eurosatnoveldatasetdeep,
      title={EuroSAT: A Novel Dataset and Deep Learning Benchmark for Land Use and Land Cover Classification}, 
      author={Patrick Helber and Benjamin Bischke and Andreas Dengel and Damian Borth},
      year={2019},
      eprint={1709.00029},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/1709.00029}, 
}