Adaptive Fuzzy Network Denoising for Enhanced Thin Ice Visualization in Cross-Polarized Sentinel-1 SAR
Year: 2026
Type: article
Abstract: Detecting thin (baby) ice in HV-polarized Sentinel-1 extra-wide (EW) sea-ice SAR images is challenging because thermal noise and the scalloping effect can mask weak backscatter signals. This paper proposes an adaptive denoising and thresholding approach using a self-designed fuzzy logic controller network (FLCN) to enhance baby-ice visualization. The approach automatically selects “no object” and “minimum object” regions and applies a data-driven correction factor to improve noise suppression without relying on external parameters. The FLCN generates input–output membership functions autonomously, reducing the need for manual tuning. We compare the results against the ESA/SNAP noise-vector correction workflow, a recent Q1 EW sea-ice denoising method [5], and a deep-learning denoiser (DnCNN). A quantitative and qualitative evaluation over multiple Sentinel-1 EW HV scenes using ROI-based open-water statistics (mean and standard deviation) and an SNR measure shows consistent noise suppression while preserving the ice–water boundaries compared to other recent methods.
Authors: Halit NAZLI, Osman YILDIRIM
Institution: Istanbul Aydın University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7138317363