Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Analytical Logit Scaling for High-Resolution Sea Ice Topology Retrieval from Weakly Labeled SAR Imagery

Reda Elwaradi, Julien Gimenez, Stéphane Hordoir, Mehdi Ait Hamma, Adrien Chan-Hon-Tong, Flora Weissgerber

Abstract

High-resolution sea ice mapping using Synthetic Aperture Radar (SAR) is crucial for Arctic navigation and climate monitoring. However, operational ice charts provide only coarse, region-level polygons (weak labels), forcing automated segmentation models to struggle with pixel-level accuracy and often yielding under-confident, blurred concentration maps. In this paper, we propose a weakly supervised deep learning pipeline that fuses Sentinel-1 SAR and AMSR-2 radiometry data using a U-Net architecture trained with a region-based loss. To overcome the severe under-confidence caused by weak labels, we introduce an Analytical Logit Scaling method applied post-inference. By dynamically calculating the temperature and bias based on the latent space percentiles (2\% and 98\%) of each scene, we force a physical binarization of the predictions. This adaptive scaling acts as a topological extractor, successfully revealing fine-grained sea ice fractures (leads) at a 40-meter resolution without requiring any manual pixel-level annotations. Our approach not only resolves local topology but also perfectly preserves regional macroscopic concentrations, achieving a 78\% accuracy on highly fragmented summer scenes, thereby bridging the gap between weakly supervised learning and high-resolution physical segmentation.

Analytical Logit Scaling for High-Resolution Sea Ice Topology Retrieval from Weakly Labeled SAR Imagery

Abstract

High-resolution sea ice mapping using Synthetic Aperture Radar (SAR) is crucial for Arctic navigation and climate monitoring. However, operational ice charts provide only coarse, region-level polygons (weak labels), forcing automated segmentation models to struggle with pixel-level accuracy and often yielding under-confident, blurred concentration maps. In this paper, we propose a weakly supervised deep learning pipeline that fuses Sentinel-1 SAR and AMSR-2 radiometry data using a U-Net architecture trained with a region-based loss. To overcome the severe under-confidence caused by weak labels, we introduce an Analytical Logit Scaling method applied post-inference. By dynamically calculating the temperature and bias based on the latent space percentiles (2\% and 98\%) of each scene, we force a physical binarization of the predictions. This adaptive scaling acts as a topological extractor, successfully revealing fine-grained sea ice fractures (leads) at a 40-meter resolution without requiring any manual pixel-level annotations. Our approach not only resolves local topology but also perfectly preserves regional macroscopic concentrations, achieving a 78\% accuracy on highly fragmented summer scenes, thereby bridging the gap between weakly supervised learning and high-resolution physical segmentation.
Paper Structure (18 sections, 3 equations, 1 figure, 1 table)

This paper contains 18 sections, 3 equations, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. Data and Problem Formulation
  3. AI4Arctic Dataset and Multimodal Fusion
  4. The Spatial Paradox of Weak Labels
  5. Methodology
  6. Base Architecture and Region-based Loss
  7. Spatial Regularization
  8. Analytical Logit Scaling
  9. Operational Large-Scale Inference
  10. Implementation and Training Details
  11. Ablation Study: Robustness of Analytical Scaling
  12. Impact of Spatial Regularization
  13. Sensitivity to Percentile Selection
  14. Experiments and Results
  15. Qualitative Evaluation: Revealing Hidden Topology
  16. ...and 3 more sections

Figures (1)

  • Figure 1: Comparison of segmentation results. (Col 1) SAR HV input showing physical topology. (Col 2) Regional weak labels from Ice Charts. (Col 3) Uncalibrated U-Net predictions exhibiting severe under-confidence. (Col 4) Our Analytical Logit Scaling revealing high-resolution leads and discrete ice floes without manual pixel-level supervision. Note: The shared colorbar represents macroscopic Sea Ice Concentration for the Ground Truth. For the calibrated prediction, the Platt Scaling pushes values to the sigmoid's saturation bounds, effectively acting as a binary topological mask.