Abstract
This thesis explores advanced deep-learning techniques for classifying hyperspectral data. Wescrutinize the background and context of hyperspectral data classification. Then, we recommend fourdeep learning architectures for this purpose. First, we introduce a two-stream spectral-spatial residualnetwork suitable for hyperspectral image (HSI) classification. Second, we propose a spatiospectralmasked autoencoder and a hybrid episode learning strategy to classify scarce HSI datasets. Third,we introduce a spectral convolution and channel attention network to capture short- and long-rangedependencies between spectral bands. Finally, we propose a transformer-based network thatincorporates a spectral-to-token module and a multiscale conformer encoder module
Original language | English |
---|---|
Qualification | Doctor of Philosophy |
Awarding Institution |
|
Supervisors/Advisors |
|
Award date | 9 May 2024 |
DOIs | |
Publication status | Unpublished - 2023 |