Adversarial Robustness of Deep Learning Models for Inland Water Body Segmentation from SAR Images

2025-05-03

AI Summary

Existing studies lack quantitative evaluation of how annotation quality—particularly adversarial label noise (e.g., label flipping, boundary perturbations)—affects the robustness of U-Net for inland water body segmentation in SAR imagery. Method: We construct the first SAR annotation noise dataset with controllable adversarial perturbations and establish a benchmark for robustness analysis. Through systematic experiments, we identify a clear robustness threshold for U-Net, quantifying the critical dependence of segmentation performance on annotation accuracy. We further propose SAR-adapted robust training strategies to mitigate noise sensitivity. Contribution/Results: Our approach significantly improves U-Net’s generalization under realistic annotation noise. We publicly release the source code, the annotated noisy dataset, and adversarial samples—establishing a new paradigm and foundational resource for trustworthy semantic segmentation in remote sensing.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Natural Language Processing: Safety and Robustness

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Semantics and Knowledge: Representation, semantic annotation, enhancement, enrichments, access and/or integration of a variety of data on the Web
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

Inland water body segmentation from Synthetic Aperture Radar (SAR) images is an important task needed for several applications, such as flood mapping. While SAR sensors capture data in all-weather conditions as high-resolution images, differentiating water and water-like surfaces from SAR images is not straightforward. Inland water bodies, such as large river basins, have complex geometry, which adds to the challenge of segmentation. U-Net is a widely used deep learning model for land-water segmentation of SAR images. In practice, manual annotation is often used to generate the corresponding water masks as ground truth. Manual annotation of the images is prone to label noise owing to data poisoning attacks, especially due to complex geometry. In this work, we simulate manual errors in the form of adversarial attacks on the U-Net model and study the robustness of the model to human errors in annotation. Our results indicate that U-Net can tolerate a certain level of corruption before its performance drops significantly. This finding highlights the crucial role that the quality of manual annotations plays in determining the effectiveness of the segmentation model. The code and the new dataset, along with adversarial examples for robust training, are publicly available. (Github link - https://github.com/GVCL/IWSeg-SAR-Poison.git)

Problem

• Assessing U-Net robustness to adversarial annotation errors in SAR water segmentation
• Studying impact of manual label noise on inland water body segmentation accuracy
• Evaluating deep learning model tolerance to data poisoning attacks in SAR imagery

Innovation

• U-Net model for SAR water segmentation
• Adversarial attacks simulate manual errors
• Public dataset and code for robust training