Residual-guided confidence learning and dynamic ensemble selection for semi-supervised hyperspectral image classification
Year: 2026
Type: article
Abstract: Hyperspectral image (HSI) classification with limited labeled samples remains challenging because unreliable pseudo-labels may propagate errors and individual classifiers may perform inconsistently across different data distributions. To address these issues, this study proposes a semi-supervised framework integrating residual-guided confidence learning and collaborative classifier strategies. Three key components are introduced. First, Co-PS selects reliable pseudo-labeled samples using confidence estimated from class-wise reconstruction residuals. Second, Co-Vote combines predictions from multiple collaborative representation classifiers to improve pseudo-label reliability. Third, Co-Select employs validation-driven classifier selection to adaptively identify a suitable classifier from the candidate pool. Experiments were conducted on four public HSI datasets using only five labeled training samples per class and ten fixed random splits. The proposed methods achieved mean OAs of up to 82.67%, 79.90%, 90.84%, and 81.47% on the Botswana, Purdue, DC, and KSC datasets, respectively, outperforming the strongest comparison baselines in mean OA. These results demonstrate the effectiveness and robustness of the proposed framework for HSI classification under extremely limited labeled supervision.
Source: Scientific Reports
Authors: Hongliang Lu, Yutian Chen, Xianglin Huang
Institutions: Tongling University, Huaiyin Normal University, Hohai University
Language: English
Subfield: Media Technology
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,690
Funder: Tongling University
OpenAlex ID: W7213646259