Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -25,9 +25,25 @@ size_categories:
|
|
25 |
This dataset contains a series of multispectral image slices captured at the embankment dams and dikes
|
26 |
of the Belo Monte Hydroelectric Complex, located in the state of Pará, northern Brazil. Each image is
|
27 |
paired with its respective NDRE vegetation index values, binary segmentation mask and multiclass
|
28 |
-
segmentation mask
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
29 |
|
30 |
-
---
|
31 |
## Dataset Structure
|
32 |
|
33 |
The dataset files are organized as following:
|
@@ -38,10 +54,28 @@ The dataset files are organized as following:
|
|
38 |
├── 📁 index_ndre # NDRE index values
|
39 |
├── 📁 labels_binary # Binary Segmentation Masks
|
40 |
├── 📁 labels_multiclass # Multiclass Segmentation Masks
|
41 |
-
└──
|
42 |
```
|
43 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
---
|
|
|
|
|
|
|
|
|
45 |
## Citation Information
|
46 |
|
47 |
If you use this dataset in your work, please cite:
|
|
|
25 |
This dataset contains a series of multispectral image slices captured at the embankment dams and dikes
|
26 |
of the Belo Monte Hydroelectric Complex, located in the state of Pará, northern Brazil. Each image is
|
27 |
paired with its respective NDRE vegetation index values, binary segmentation mask and multiclass
|
28 |
+
segmentation mask.
|
29 |
+
|
30 |
+
The multispectral images were captured by the Micasense RedEdge-P multispectral sensor embedded in a
|
31 |
+
DJI M210 V2 UAV. Radiometric calibration was performed for all images based on the known reflectance
|
32 |
+
values of a calibration panel. All images were used to process a Digital Ortophoto Map, which was then
|
33 |
+
sliced into the 256x256x5 image patches that are contained in this dataset.
|
34 |
+
|
35 |
+
Each image file is composed of six channels:
|
36 |
+
|
37 |
+
1. Red band reflectance
|
38 |
+
2. Green band reflectance
|
39 |
+
3. Blue band reflectance
|
40 |
+
4. Red Edge band reflectance
|
41 |
+
5. Near-infrared band reflectance
|
42 |
+
6. Binary cutline
|
43 |
+
|
44 |
+
The vegetation index (NDRE) values were calculated based on the spectral bands and the segmentation masks
|
45 |
+
were manually annotated using the CVAT software.
|
46 |
|
|
|
47 |
## Dataset Structure
|
48 |
|
49 |
The dataset files are organized as following:
|
|
|
54 |
├── 📁 index_ndre # NDRE index values
|
55 |
├── 📁 labels_binary # Binary Segmentation Masks
|
56 |
├── 📁 labels_multiclass # Multiclass Segmentation Masks
|
57 |
+
└── 📝 README.md
|
58 |
```
|
59 |
|
60 |
+
## Segmentation Classes
|
61 |
+
|
62 |
+
The image annotations are formatted for both binary and multi-class segmentation.
|
63 |
+
|
64 |
+
**Binary Segmentation Classes:**
|
65 |
+
- Slope
|
66 |
+
- Not-Slope
|
67 |
+
|
68 |
+
**Multi-class Segmentation Classes:**
|
69 |
+
- Slope
|
70 |
+
- Drainage Channels
|
71 |
+
- Stairways
|
72 |
+
- Background
|
73 |
+
|
74 |
---
|
75 |
+
## License Information
|
76 |
+
|
77 |
+
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.
|
78 |
+
|
79 |
## Citation Information
|
80 |
|
81 |
If you use this dataset in your work, please cite:
|