Infrared hyperspectral data are typical meteorological observations, which can detect the atmosphere vertically in many spectrums. Distinguishing obsorption peaks of different materials appearing in specific spectrums can help classify those materials. However, there are three characteristics of these data, namely:
- high spectral correlation,
- high spatial correlation,
- and sparsity,
and they casue a trouble during processing. This talk explained why they are highly correlated but also sparse. Kernel PCA for compressing was also tested.
Check slides for more details.