Creation of BigEarthNet-Sen1

---

This repository contains the data processing codes for Sentinel-1 VVVH image creation of BigEarthNet dataset. For the details about BigEarthNet, please kindly refer to this website. We note the original BigEarthNet dataset as BigEarthNet-Sen2 hereafter.

Requirements

• SNAP Toolbox

• Sen1mosaic python toolbox

Pipeline

Retrieve the Sentinel-1 tiles for downloading

Retrieve the geo-locations of the Sentinel-2 tiles which are utilized for the creation of BigEarthNet-Sen2. These Sentinel-2 tiles can be obtained through this link.

• Getting the associated footprints of Sentinel-2 tiles from the meta data.
• Retrieve the corresponding Sentinel-1 tiles based on the footprints.
• For each Sentinel-2 tile, there may be several Sentinel-1 tiles covering it. In order to significantly reduce the number of Sentinel-1 tiles to be downloaded, relative orbit of each tile should be retrieved. For each relative orbit, we just keep one Sentinel-1 tile, to avoid the duplicated Sentinel-1 tiles for downloading.
• Also, we should make sure that the union of the footprints of the Sentinel-1 tiles should cover the assocaited Sentinel-2 tile.
• Finally, we can get the product information of the Sentinel-1 tiles to be downloaded.

After this step, we get a npy file BigEarthNetTileInfo_v2.npy which contains the basic information of Sentinel-2 tiles utilized in BigEarthNet-Sen2. Also, we have BigEarthNetSen2UUID2Sen1ProdName.npy, which links the Sentinel-2 tiles with respect to the assocaited Sentinel-1 tiles to be downloaded. We construct the npy file as a nested dictionary.

Loading BigEarthNetSen2UUID2Sen1ProdName.npy via:

sen1_tile_download_info_pth = './BigEarthNetSen2UUID2Sen1ProdName.npy'
data = np.load(sen1_tile_download_info_pth, allow_pickle=True).item()
Sen2UUID_2_Sen1ProdNames = data['Sen2UUID_2_Sen1ProdName']

Loading BigEarthNetTileInfo_v2.npy via:

sen2_tile_info_pth = './BigEarthNetTileInfo_v2.npy'
data = np.load(sen2_tile_info_pth, allow_pickle=True).item()
tile_info = data['tile_info']

The keys of tile_info are the tiling grids of the Sentinel-2 tiles of BigEarthNet-Sen2. Within each tiling grid, the keys represent the UUID of the Sentinel-2 tiles. Then, we retrieve the Sentinel-1 tiles accordingly.

for tile_grid in list(tile_info.keys()):
    for uuid in list(tile_info[tile_grid].keys()):
        sen1_tiles = Sen2UUID_2_Sen1ProdNames[uuid]

Then, we can download the assocaited Sentinel-1 tiles.

Downloading Sentinel-1 tiles

We download the corresponding tiles via ./sen1mosaic_scripts/down_sen1_tiles.py

Preprocessing the downloaded Sentinel-1 tiles

Preprocess the downloaded Sentinel-1 tiles based on the following steps:

• GRD border noise removal

• Thermal noise removal

• Radiometric calibration

• Speckle noise filtering

• Terrain correction

• Transform into decibels (dB)

The associated scripts can be found in: ./sen1mosaic_scripts/proc_sen1_tiles.py

Please note that in the step of Terrain correction, for some tiles with high latitude, ASTER DEM is needed, SRTM is out of the coverage. Therefore, the corresponding xml file should be modified. For reference, refer to Sentinel-1 preprocessing on Google Earth Engine.

Mosaicing the processed SAR images

The script of this step is ./sen1mosaic_scripts/mosaic_sen1_tiles.py

Clipping the SAR images

Clip the mosaic SAR images with respect to the associated Sentinel-2 tiles via ./sen1mosaic_scripts/clip_patches_sen1_tiles.py

At last, we can obtain the BigEarthNet-Sen1 dataset based on the names of BigEarthNet-Sen2.

Visualization

Left is the Sentinel-2 image, Right is the RGB visualization of Sentinel-1 VVVH bands

Acknowledgement

Some discussions are done in this work, hereby, I would like to thank:

• Gencer Sümbül
• Gerald Baier
• Samuel Bowers
• Jingliang Hu