This document is part of the RepreSent project proposal that was accepted in January 2022 and responded to ESA ITT for AI4EO Challenges – Non-Supervised Learning (AO/1-10552/21/I-DT). The project is performed under ESA Contract No. 4000137253/22/I-DT.
The main scope of the RepreSent project is to design, implement and validate artificial intelligence (AI) non-supervised techniques that will allow the use of the Copernicus Sentinel data. These techniques are developed for:
- Fusion of Sentinel sensors (e.g., Sentinel-1 and Sentinel-2)
- Single sensor classification (multispectral or SAR Sentinel sensors)
- Change detection (e.g., Sentinel-1 or Sentinel-2)
- Image time series analysis (e.g., Sentinel-2)
The software is implemented as an open source library. The validation is done on five use cases (UC) related to:
- UC1) Forest disturbance monitoring
- UC2) Automated Land Cover mapping
- UC3) Anomaly detection in long time series of PS-P InSAR
- UC4) Cloud detection and removal
- UC5) Forest biomass estimation
The resulting datasets of the project is to be distributed freely to the AI4EO community.
The dataset provided as part of the RepreSent project follows a specific folder structure to organize the files and code. Here's an overview of the directory structure:
-
notebooks
-
represent
- callbacks
- config
- data
- datamodules
- experiments
- uc1_contrastive_learning
- uc1_forest_change_map
- uc2_settlement_evaluation.py
- uc3_building_anomaly_detection
- uc3_benchmark
- ganf
- maxdiv
- dense_autoencoder.py
- uc3_lstm_autoencoder.py
- uc4_odc.py
- uc4_resnet.py
- losses
- models
- moco.py
- simclr_resnet.py
- uc1_byol.py
- uc1_pixel_level_contrastive_learning.py
- uc1_resnet_base.py
- uc1_resnet_dcva.py
- uc2_maml.py
- uc2_segmentation_resnet.py
- uc2_supervised_resnet.py
- uc3_benchmark
- ganf
- maxdiv
- uc3_lstm_autoencoder.py
- uc4_odc.py
- uc4_resnet.py
- tools
The trained weights can be found obtained from HERE. Unzip the weights into the code
represent
directory for using them with the notebooks.