Urban monthly land dynamics Sentinel-2 benchmark dataset

The public data set of the paper "Time-series land cover change detection using deep learning-based temporal semantic segmentation" uses monthly synthesized Sentinel-2 for time series semantic change detection. A total of 32894 samples were collected. Each timestamp has a land cover type annotation. Anyone can use this data set to conduct further research. We will add more areas in the future. 

Data description:

The time series length of the sample is 48, 48 months.

The dimension of the time series is 13: [Sentinel-2 data (10) + Lable (1) + Lon(1) + Lat (1) ].

Label mapping: 0 is water body, 1 is woodland, 2 is grassland, 3 is bare soil, 4 is impervious surface, 5 is cropland.

For any implementation details, you can refer to the paper or github.

Paper citations:

He H, Yan J, Liang D, Sun Z, Li J, Wang L. Time-series land cover change detection using deep learning-based temporal semantic segmentation. Remote Sensing of Environment. 2024, 305:114101.