README.md

---
license:
- unknown
task_categories:
- image-classification
language:
- en
tags:
- remote-sensing
- earth-observation
- geospatial
- satellite-imagery
- scene-classification
pretty_name: RESISC45 Dataset
size_categories:
- n<1G
---

# Remote Sensing Image Scene Classification (RESISC45) Dataset

- **Paper** [Remote Sensing Image Scene Classification: Benchmark and State of the Art
](https://arxiv.org/pdf/1703.00121.pdf)
- **Paper with code**: [RESISC45](https://paperswithcode.com/dataset/resisc45)

![RESISC45](./thumbnail.png)

## Description

The RESISC45 dataset is a scene classification dataset that focuses on RGB images extracted using [Google Earth](https://earth.google.com/web/). This dataset comprises a total of 31,500 images, with each image having a resolution of 256x256 pixels. RESISC45 contains 45 different scene classes, with 700 images per class. These images are collected from over 100 countries and were specifically selected to optimize for high variability in image conditions, including spatial resolution, occlusion, weather, illumination, and more. Among its notable features, RESISC45 contains varying spatial resolution ranging from 20cm to more than 30m/px.

## Details

## Structure

```tree
.
├── README.md
└── data
   ├── airplane
   │   ├── airplane_1.jpg
   │   ├── ...
   │   └── airplane_700.jpg
   ├── airport
   ├── baseball_diamond
   ├── beach
   ├── ...
   └── wetland
```

### Statistics
- Total Number of Images: 31,500
- Image Resolution: 256x256 pixels
- Scene Categories: 45
- Dataset Size: Approximately 0.47GB

## Citation

If you use the RESISC45 dataset in your research, please consider citing the following publication or the dataset's official website:

```bibtex
@article{cheng2017remote,
  title     = {Remote sensing image scene classification: Benchmark and state of the art},
  author    = {Cheng, Gong and Han, Junwei and Lu, Xiaoqiang},
  journal   = {Proceedings of the IEEE},
  volume    = {105},
  number    = {10},
  pages     = {1865-1883},
  year      = {2017},
  publisher = {IEEE}
}
```