README.md

---
license: cc-by-nc-sa-4.0
task_categories:
- image-segmentation
tags:
- remote-sensing
- uav
- multispectral
- land-cover
- segmentation
- dam
- dike
- slope
- vegetation
pretty_name: Dam Segmentation Dataset
size_categories:
- 1K<n<10K
---

# Dam Segmentation Dataset
# Multispectral UAV Remote Sensing Data for Embankment Dam Segmentation

## Dataset Summary

This dataset contains a series of multispectral image slices captured at the embankment dams and dikes 
of the Belo Monte Hydroelectric Complex, located in the state of Pará, northern Brazil. Each image is 
paired with its respective NDRE vegetation index values, binary segmentation mask and multiclass 
segmentation mask.

The multispectral images were captured by the Micasense RedEdge-P multispectral sensor embedded in a 
DJI M210 V2 UAV. Radiometric calibration was performed for all images based on the known reflectance
values of a calibration panel. All images were used to process a Digital Ortophoto Map, which was then 
sliced into the 256x256x6 image patches that are contained in this dataset. Regions from each of the 
ortophotos were assigned for training (70%), validation (15%) and testing (15%).

Each image file is composed of six channels:

1. Red band reflectance
2. Green band reflectance
3. Blue band reflectance
4. Red Edge band reflectance
5. Near-infrared band reflectance
6. Binary cutline

The vegetation index (NDRE) values were calculated based on the spectral bands and the segmentation masks
were manually annotated using the CVAT software.

## Dataset Structure

The dataset files are organized as following:

```
📁 dam-segmentation
├── 📁 test                         # Test dataset
│   ├── 📁 images
│   ├── 📁 mask_binary
│   └── 📁 mask_multiclass
├── 📁 train                       # Training dataset
│   ├── 📁 images
│   ├── 📁 mask_binary
│   └── 📁 mask_multiclass
├── 📁 val                         # Validation dataset
│   ├── 📁 images
│   ├── 📁 mask_binary
│   └── 📁 mask_multiclass
└── 📝 README.md

```

## Segmentation Classes

The image annotations are formatted for both binary and multi-class segmentation.

**Binary Segmentation Classes:**
- Slope
- Not-Slope

**Multi-class Segmentation Classes:**
- Slope
- Drainage Channels
- Stairways
- Background

---
## License Information

This dataset is licensed under the [Creative Commons Attribution Non Commercial Share Alike 4.0 International](https://spdx.org/licenses/CC-BY-NC-SA-4.0) license terms.

## Citation Information

If you use this dataset in your work, please cite:

```latex
@misc{teixeira2024damseg,
    author = {Carlos André de Mattos Teixeira},
    title = {Multispectral UAV Remote Sensing Data for Embankment Dam Segmentation},
    year = {2024},
    publisher = {Hugging Face},
    journal = {Dataset Repository},
    url = {https://huggingface.co/datasets/andrematte/dam-segmentation}
    doi = { 10.57967/hf/3089 },
}
```