Skip to main navigation Skip to search Skip to main content

Hybrid Transformer-CNN with Two-Stage Training for SAR Despeckling

  • Zhenyu Guo
  • , Weidong Hu
  • , Shichao Zheng
  • , Yuanyuan Zhang
  • , Ming Zhou
  • , Qian Wang
  • , Ruizhe Zhang
  • Beijing Institute of Technology
  • China Aerospace Science and Technology Corporation
  • Jiaxing Glead Electronics Co., Ltd.

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Coherent speckle noise in Synthetic Aperture Radar (SAR) imagery severely degrades image fidelity, compromising the accuracy and reliability of subsequent advanced interpretation tasks such as target recognition, scene classification, and semantic segmentation. Convolutional neural networks (CNNs) face a significant domain gap due to their reliance on synthetic data, resulting in insufficient generalization capabilities in real-world scenarios. To address this challenge, we propose a novel denoising framework centered on a hybrid network combining hierarchical Transformers and CNNs. This framework bridges the domain gap through a two-stage 'pre-training-fine-tuning' paradigm and achieves end-to-end optimization via a spatiotemporalfrequency domain joint loss function. Experiments on synthetic and real SAR datasets demonstrate that our approach achieves superior performance across multiple key metrics and visual quality, significantly enhancing despeckling capabilities and detail fidelity.

Original languageEnglish
Title of host publicationCISS 2025 - 6th China International SAR Symposium
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9798319517609
DOIs
Publication statusPublished - 2025
Externally publishedYes
Event6th China International SAR Symposium, CISS 2025 - Yiwu, China
Duration: 25 Oct 202527 Oct 2025

Publication series

NameCISS 2025 - 6th China International SAR Symposium

Conference

Conference6th China International SAR Symposium, CISS 2025
Country/TerritoryChina
CityYiwu
Period25/10/2527/10/25

Keywords

  • Despeckling
  • Domain Gap
  • Multitemporal approach
  • Synthetic Aperture Radar
  • Transformers

Fingerprint

Dive into the research topics of 'Hybrid Transformer-CNN with Two-Stage Training for SAR Despeckling'. Together they form a unique fingerprint.

Cite this