NSegment : Noisy Segment Improves Remote Sensing Image Segmentation

2025-04-28

AI Summary

Remote sensing image segmentation faces dual challenges: implicit label noise—including ambiguous boundaries, mixed pixels, shadows, and annotator subjectivity—and severe scarcity of labeled data. To address these, we propose a lightweight, end-to-end data augmentation method that applies dynamic-intensity elastic deformation exclusively to segmentation labels for explicit noise modeling—requiring neither label correction nor sample filtering. This is the first work to confine elastic deformation to the label space and adapt deformation intensity dynamically via sampling to match varying noise levels, thereby enhancing model robustness without compromising training efficiency. Extensive experiments across multiple state-of-the-art segmentation architectures demonstrate consistent and significant accuracy improvements, effectively mitigating the impact of implicit label noise. Crucially, the method introduces negligible computational overhead, making it highly practical for real-world remote sensing applications.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Intelligent Robots: Localization, Mapping, and Navigation

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Security and Privacy: Data transparency and provenance

Abstract

Labeling errors in remote sensing (RS) image segmentation datasets often remain implicit and subtle due to ambiguous class boundaries, mixed pixels, shadows, complex terrain features, and subjective annotator bias. Furthermore, the scarcity of annotated RS data due to high image acquisition and labeling costs complicates training noise-robust models. While sophisticated mechanisms such as label selection or noise correction might address this issue, they tend to increase training time and add implementation complexity. In this letter, we propose NSegment-a simple yet effective data augmentation solution to mitigate this issue. Unlike traditional methods, it applies elastic transformations only to segmentation labels, varying deformation intensity per sample in each training epoch to address annotation inconsistencies. Experimental results demonstrate that our approach improves the performance of RS image segmentation on various state-of-the-art models.

Problem

• Addresses labeling errors in remote sensing image segmentation
• Mitigates scarcity of annotated remote sensing data
• Reduces training complexity for noise-robust models

Innovation

• Elastic transformations on segmentation labels
• Varying deformation intensity per sample
• Simple effective data augmentation solution