Despeckling Polarimetric SAR Data Using a Multistream Complex-Valued Fully Convolutional Network

A polarimetric synthetic aperture radar (PolSAR) sensor is able to collect images in different polarization states, making it a rich source of information for target characterization. PolSAR are inherently affected by speckle. Therefore, before deriving ad hoc products from the data, covariance matrix needs be estimated reducing In recent years, deep learning-based despeckling methods have started evolve single-channel SAR images. To this aim, approaches separate real and imaginary components complex-valued use them as independent channels standard convolutional neural networks (CNNs). However, approach neglects mathematical relationship that exists between components, resulting suboptimal output. Here, we propose multistream fully network (FCN) (CV-deSpeckNet https://github.com/adugnag/CV-deSpeckNet ) reduce speckle effectively estimate matrix. evaluate performance CV-deSpeckNet, used Sentinel-1 dual compare against its real-valued counterpart separates parts complex CV-deSpeckNet was also compared state art methods. The results show trained with fewer number samples, has higher generalization capability, resulted accuracy than state-of-the-art These showcase potential learning despeckling.