This dataset was created from Gaofen-2 satellite imagery covering Shenzhen, China. It includes scene level sample sets of built-up and non built-up areas, as well as pixel level annotated test sets.
- Image Channels: Three channels of RGB with panchromatic sharpening
- Spatial Resolution: 1 m
- Total number of images: 9808
- Image size: 112 × 112
| Built-up Areas | Non-built-up Areas | |
|---|---|---|
| Manually Selected Sample Count | 1226 | 1226 |
| Sample count obtained through data augmentation | 3678 | 3678 |
| Total Sample Count | 4904 | 4904 |
Note: During the data augmentation process, image flipping was adopted, with 1226 selected built-up and non built-up samples flipped at 90, 180, and 270 degrees, resulting in a fourfold increase in sample data.
The test set includes satellite images of five sub-regions in Shenzhen, along with corresponding ground truth annotation data.The specific sizes are as follows:
- Test 1: 8960 × 7840
- Test 2: 10,080 × 8960
- Test 3: 8400 × 8400
- Test 4: 8960 × 8960
- Test 5: 8400 × 8400
This dataset may provide necessary data reference and support for researchers in the detection, segmentation, and mapping of urban built-up areas. Previous studies related to this dataset can refer to:
[1]Chen, Y.; Peng, F.; Yao, S.; Xie, Y. Lightweight Multilevel Feature- Fusion Network for Built-Up Area Mapping from Gaofen-2 Satellite Images. Remote Sens. 2024, 16.
[2]Chen, Y.; Yao, S.; Hu, Z.; Huang, B.; Miao, L.; Zhang, J. Built-up Area Extraction Combing Densely Connected Du-al-Attention Network and Multi-Scale Context. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 2023.
If you have any questions or suggestions, please contact us：chenyixiang@njupt.edu.cn; 1022173204@njupt.edu.cn
link：https://pan.baidu.com/s/1NJXt4BakUWi9zdxMhhUT6Q?pwd=66gg Extraction Password：66gg