LandCoverPT Dataset

LandCoverPT - A dataset for training Machine Learning models to classify the Portuguese territory land cover.
The creation of the dataset used 26 Sentinel-2 products, captured in June and August 2019, and the products where divided into 153347 patches with 120×120 pixels each. The products cover the Portuguese mainland. The Sentinel-2 data was complemented with Corine Land Cover 2018 data.
Each patch includes B01, B02, B03, B04, B05, B06, B07, B08, B8A, B09, B11, and B12 Sentinel-2 bands, and the CLC 2018 layer.
The dataset is stored in 3 TFRecord files, for training, validation, and test. The training set contains 98141 patches, the validation set includes 24536 patches, and the test set contains 30670 patches.
It is also provided code files to experiment with the dataset:
- two configuration files in JSON format
- a python file with the U-Net model
- a python file with a class to create a tf.data.Dataset based on the TFRecord files
- a notebook for training the U-Net model on the LandCoverPT dataset
- a notebook for evaluating the U-Net model trained on the LandCoverPT dataset
- a notebook to make predictions with the U-Net model trained on the LandCoverPT dataset.