EUVIMEDEUVIMED entdecken
Crossref

MLDGNet: MSB-LSB Guided Dual-Branch Network With Anisotropic Refinement for SAR Despeckling

Tanveer Fatema Khan, Ramesh Kumar Thakur, Manasi Gyanchandani

IEEE Signal Processing Letters · 2026 · 33 · 4286-4290

DOI: 10.1109/lsp.2026.3734820 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Signal Processing Letters, 2026. Verfasst von Tanveer Fatema Khan und 2 weiteren Autor:innen. Die Arbeit nennt 20 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Speckle noise is a persistent problem in Synthetic Aperture Radar(SAR) images because it mixes with the signal in a multiplicative way and often hides important structures. To address this problem more successfully, this letter proposed a bit-plane slicing concept together with an edge-preserving anisotropic diffusion refinement stage. The Most Significant Bits(MSB) and Least Significant Bits(LSB) planes of the noisy image are first separated, and as they contain distinct types of information, each half is processed using a separate network. The sharpening model used in the MSB path primarily enhances edges and larger structural elements. A tiny pixel-attention module is used in the LSB path to reduce the unstable noise patterns that typically occur in lower bits while attempting to preserve relevant fine textures. An intermediate result with a clearer structure than the input is created by combining the outputs of the two branches after they have been trained. The remaining speckle is then cleaned using a anisotropic diffusion module without erasing the sharp details that were previously retrieved. The final SAR image has improved edge sharpness and more dependable texture restoration because of this combination arrangement.

Zitieren

APA

Khan, T. F., Thakur, R. K., & Gyanchandani, M. (2026). MLDGNet: MSB-LSB Guided Dual-Branch Network With Anisotropic Refinement for SAR Despeckling. IEEE Signal Processing Letters, 33, 4286-4290. https://doi.org/10.1109/lsp.2026.3734820

Vancouver

Khan TF, Thakur RK, Gyanchandani M. MLDGNet: MSB-LSB Guided Dual-Branch Network With Anisotropic Refinement for SAR Despeckling. IEEE Signal Processing Letters. 2026;33:4286-4290. doi:10.1109/lsp.2026.3734820

BibTeX
@article{Khan2026,
  author = {Khan, Tanveer Fatema and Thakur, Ramesh Kumar and Gyanchandani, Manasi},
  title = {MLDGNet: MSB-LSB Guided Dual-Branch Network With Anisotropic Refinement for SAR Despeckling},
  journal = {IEEE Signal Processing Letters},
  year = {2026},
  volume = {33},
  pages = {4286--4290},
  doi = {10.1109/lsp.2026.3734820}
}
Kostenlos starten →