Abstract:
This thesis presents a novel paradigm to analyze the remotely sensed hyperspectral imagery, i.e. hyperspectral dimensionality reduction, spectral unmixing, cross-modality fusion and learning. The trade-off between spectral robustness and discrimination is considered by regression-based representation models. Accordingly, the optimizers are designed for the solutions of these models. Results are assessed on several datasets in comparison with state-of-the-art methods.