Cardiff University | Prifysgol Caerdydd ORCA
Online Research @ Cardiff 
WelshClear Cookie - decide language by browser settings

Integrating semi-supervised and active learning for semantic segmentation

Ma, Wanli, Karakus, Oktay ORCID: https://orcid.org/0000-0001-8009-9319 and Rosin, Paul L. ORCID: https://orcid.org/0000-0002-4965-3884 2026. Integrating semi-supervised and active learning for semantic segmentation. Science of Remote Sensing 13 , 100427. 10.1016/j.srs.2026.100427

[thumbnail of 1-s2.0-S2666017226000659-main.pdf] PDF - Published Version
Available under License Creative Commons Attribution.

Download (4MB)

Abstract

Pixel-level annotation for image segmentation tasks is both time-consuming and expensive, particularly in the context of remote sensing. To mitigate this challenge, semi-supervised learning and active learning offer effective solutions, although they are typically employed independently. In this paper, we propose a novel hybrid learning framework that integrates active learning with an enhanced semi-supervised learning strategy, tailored to reduce annotation costs and improve semantic segmentation performance, particularly in the domain of remote sensing, where dense pixel-level annotation is expensive and labour-intensive. Our method leverages both the labelled samples selected via active learning and the unlabelled data excluded from selection by incorporating a pseudo-label auto-refinement (PLAR) mechanism. This module identifies potentially inaccurate pseudo-labels and automatically refines them based on the cluster assumption that similar features in high-density regions of the feature space correspond to the same class. Importantly, manual annotation is limited to only the most uncertain regions, while less ambiguous areas are refined without consuming labelling budget. We evaluate our proposed framework on multiple benchmark datasets covering both natural and remote sensing imagery. The results demonstrate that our method consistently outperforms state-of-the-art baselines in image segmentation tasks, confirming its effectiveness and generalisation capability across different domains.

Item Type: Article
Date Type: Publication
Status: Published
Schools: Schools > Computer Science & Informatics
Publisher: Elsevier BV
ISSN: 2666-0172
Date of First Compliant Deposit: 20 April 2026
Date of Acceptance: 7 April 2026
Last Modified: 25 Aug 2026 22:09
URI: https://orca.cardiff.ac.uk/id/eprint/186505

Actions (repository staff only)

Edit Item Edit Item

Downloads

Downloads per month over past year

View more statistics