Multisensor Hyperspectral Image Classification via Denoising-CAE and Dataset-Aware MAE
IEEE Geoscience and Remote Sensing Letters · 2026 · 23 · 5508005-5508005
DOI: 10.1109/lgrs.2026.3743370 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Geoscience and Remote Sensing Letters, 2026. Verfasst von Xinyi Lin und 1 weiteren Autor:innen. Die Arbeit nennt 19 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Natural Science Foundation of Heilongjiang Province of China.
Abstract (Originalsprache)
Hyperspectral image (HSI) classification has achieved remarkable performance with deep learning methods. However, due to the varying spectral dimensionalities of HSIs acquired by different sensors, existing models are generally limited to single-dataset representation learning. As the number of HSI sensors continues to grow, this limitation restricts practical applications. To address this issue, we propose a novel framework for multi-sensor HSI classification via denoising convolutional autoencoder and dataset-aware masked autoencoder and dataset-aware masked autoencoder (MuSD). MuSD consists of two key components: a denoising convolutional autoencoder (denoising-CAE) for spectral dimension unification and a dataset-aware masked autoencoder (dataset-aware MAE) with dataset identity (DID) embedding for joint multisensor representation learning. In particular, the denoising-CAE reconstructs HSIs from different sensors and generates feature representations with unified spectral dimensionality, while noise injection is adopted to improve feature robustness. The unified features are then jointly utilized for MAE-based self-supervised pretraining with DID, which alleviates feature confusion across datasets and enables the model to learn generalized spectral–spatial representations from unlabeled samples. Consequently, the dataset-aware MAE bypasses the inconsistency of category annotations and effectively exploits large-scale HSI data from multiple sensors. Extensive experiments on real HSI datasets demonstrate that the proposed method consistently outperforms conventional single-dataset training approaches and achieves competitive classification performance across multiple sensors. The code will be available at https://github.com/LinXY82586/MuSD
Förderung
- National Natural Science Foundation of China
- Natural Science Foundation of Heilongjiang Province of China
Zitieren
APA
Lin, X., & Chen, Y. (2026). Multisensor Hyperspectral Image Classification via Denoising-CAE and Dataset-Aware MAE. IEEE Geoscience and Remote Sensing Letters, 23, 5508005-5508005. https://doi.org/10.1109/lgrs.2026.3743370
Vancouver
Lin X, Chen Y. Multisensor Hyperspectral Image Classification via Denoising-CAE and Dataset-Aware MAE. IEEE Geoscience and Remote Sensing Letters. 2026;23:5508005-5508005. doi:10.1109/lgrs.2026.3743370
BibTeX
@article{Lin2026,
author = {Lin, Xinyi and Chen, Yushi},
title = {Multisensor Hyperspectral Image Classification via Denoising-CAE and Dataset-Aware MAE},
journal = {IEEE Geoscience and Remote Sensing Letters},
year = {2026},
volume = {23},
pages = {5508005--5508005},
doi = {10.1109/lgrs.2026.3743370}
}
EUVIMED