Deep Learning for InSAR Phase Filtering: An Optimized Framework for Phase Unwrapping

Interferometric Synthetic Aperture Radar (InSAR) data processing applications, such as deformation monitoring and topographic mapping, require an interferometric phase filtering step. Indeed, the quality significantly impacts terrain height estimation accuracy. However, existing classical deep learning-based methods provide artefacts in filtered areas where a large amount of noise prevents retrieving original signal. In this way, we can no longer distinguish underlying informative signal for next This paper proposes convolutional neural network method, developing novel learning strategy to preserve initial input into these crucial areas. Thanks encoder–decoder powerful feature extraction ability, predict accurate coherence phase, ensuring reliable final results. Furthermore, also address (SAR) interferograms simulation that, using parameters estimated from real SAR images, considers physical behaviors typical acquisition. According results achieved on simulated InSAR data, proposed method outperforms ones.