An Improved Least Square Phase Unwrapping Algorithm Combined with Convolutional Neural Network

Abstract

Phase unwrapping (PU) technology plays a decisive role in interferometric synthetic aperture radar (InSAR) workflow. Due to the limitation of InSAR system, the interferometric phase can only be measured between (-π, π), which called wrapped phase. In order to overcome this limitation and obtain the absolute phase, phase unwrapping algorithm came into being. The least square phase unwrapping algorithm is one of the most popular phase unwrapping methods, which can be solved by fast Fourier transform (FFT) and has high efficiency. It means that this algorithm can waste less time to obtain the InSAR products, while its accuracy is often unstable. This paper is dedicated to overcoming this instability through combining deep learning with fast least square method. The specific steps include: 1. Use deep learning to predict the phase gradient stably regardless of the phase quality; 2. Replace the wrapped phase gradient of least square unwrapping algorithm by above prediction result and take the least square solution. The simulation and real data experimental results demonstrate the effectiveness of this improved method.