SITS-Former: A pre-trained spatio-spectral-temporal representation model for Sentinel-2 time series classifcation (original) (raw)
Published December 24, 2021 | Version v1
Dataset Open
- 1. Nanjing University of Posts & Telecommunications
Description
This is the unlabeled dataset we introduced in the presented paper 'SITS-Former: A pre-trained spatio-spectral-temporal representation model for Sentinel-2 time series classifcation'. This dataset can be used to pre-train a specified deep learning model (such as SITS-Former, CNN-Transformer, ConvLSTM. etc) for patch-based Sentinel-2 time series classification.
In this dataset, each sample corresponds to an unlabeled image patch time series, which is stored as a separate numpy file named 'unlabeled_XXX.npz'. You can use 'np.load' to open a saved '.npz' file and get two arrays (querid by "ts" and "doy") from the returned dictionary. The code will be released at https://github.com/linlei1214/SITS-Former soon.
Files
Files (15.4 GB)