Phase Unwrapping using ML methods

• Autor(a) principal: Couto, Diogo Gabriel da Silva
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10348/11749
• topic: InSAR; SNAPHU

Interferometric Synthetic Aperture Radar (InSAR) is an important technology that has revolutionized the way we study the Earth's surface, allowing us to measure tiny changes in the ground surface with high precision and accuracy. One of the biggest challenges in InSAR is the phase unwrapping process, which involves removing the ambiguity in the measured phase caused by the periodicity of the radar signal. This computationally intensive and time-consuming task requires sophisticated algorithms and significant computational resources. The new deep learning algorithms, such as Generative Adversarial Networks (GANs), may be an alternative to simplify the phase unwrapping algorithm. In this work, we evaluate GANs for InSAR phase unwrapping. The proposed method replaces Statistical-Cost, Network-Flow Algorithm for Phase Unwrapping (SNAPHU) with GANs, which can achieve considerably faster processing times with little degradation in quality. The GANs can unwrap each phase in about 30 seconds, while SNAPHU can take up to 1 minute and 15 seconds. A difference map comparison between the Small Baseline Subset (SBAS) results using GANs and SNAPHU unwrapped phases shows that approximately 84% of the GANs points are within 3 millimetres of SNAPHU. The attained result of GANs represents a significant step forward in the evolution of phase unwrapping methods. While this experiment did not show a definitive winner due to the restrictive nature of the tests, it demonstrates that GANs are a viable alternative in some instances and may replace SNAPHU as the preferred unwrapping method.