Unsupervised learning for synthetic aperture radar image despeckling
Addressing the challenge of traditional deep learning-based SAR image despeckling methods requiring a large amount of annotated data for supervised training,this paper proposes a novel unsupervised SAR image despeckling approach.Leveraging the capability of cycle-consistent generative adversarial networks,the despeckling problem is transformed into an unpaired image-to-image translation task,thus bypassing the reliance on ground truth labeled datas.Experimental results demonstrate that this method is not only effective in suppressing speckle in SAR images but also capable of learning to generate realistic SAR image speckle with one-to-many characteristics,without dependence on data simulation or assumptions about the speckle distribution.The proposed approach holds promising potential applications in the field of SAR image processing,offering an effective and unsupervised learning-based solution to SAR image despeckling.
NETL
NSTL
万方数据
