UNet Based Deep Learning for Interferometric SAR Phase Unwrapping

초록

Phase Unwrapping (PU) is a fundamental task in Interferometric Synthetic Aperture Radar (InSAR) processing. conventional model-based approaches often struggle in noisy or complex fringe environments. In this work, we propose a convolution layer-based deep learning model that directly regresses the unwrapped phase from the single-channel-wrapped input. By formulating PU as a supervised regression problem and training with a synthetic interferogram, the proposed method achieves a low root mean square error without relying on explicit residue or branch-cut heuristics. The experimental results demonstrate the robustness and precision of the model, highlighting the potential of learning-based methods in InSAR phase recovery tasks.

키워드