AIDA Cropland delination model
Use cases
it takes a Sentinel-2 image time series as input and predicts a segmentation map of 6 crop classes. It is trained on a Sentinel-2 images of Germany, Lithuania, Latvia, Estonia, Belgium, Austria, Slovenia, Slovakia and Netherlands tagged using Eurocrops
Model
The model used is a modified version of UTAE, a U-Net like architecture which consists of a convolutional encoder-decoder and central attention module.
Input
All pre-trained models expect input images normalized in the same way, i.e. mini-batches of 10-channel Sentinel-2 images of shape (B x 6 x 10 x 128 x 128), where B is the batch size and the second dimension is the 6 months used for prediction (April to September).
Preprocessing
The images have to be divided by a factor of 10000 and then clipped in a range of [0, 1].
Output
The model outputs a segmentation map of the input image.
int[1, 1, 128, 128]
Contributors
- gaetanochiriaco (LINKS Foundation)
- edornd (LINKS Foundation)
License
MIT License