Spatially Aligned Domain-Adversarial Neural Networks with self-training
This repository provides code for SpADANN framework which allows temporal domain adaptation (temporal transfer) of classification model designed to produce land cover maps based on Satellite Image Time Series. The code uses Python and TensorFlow 2.X with Keras as the main model building API.
This framework is detailed in the research paper "Temporal Domain Adaptation for Satellite Image Time Series Land Cover Mapping with Adversarial Learning and spatially-aware Self-Training" which is currently under submission.
The explanations regarding the format of SITS data handled by SpADANN are included in Python script "main_SpADANN.py".
This project is given under the terms of the MIT license.