Hyperspectral image analysis

The new generation of earth observation sensors include hyperspectral sensors with a high number of spectral bands. Each pixel in a hyperspectral image with D spectral bands will result in a D-dimensional vector corresponding to the measured energy in dif ferent parts of the electromagnetic spectrum. Although the higher spectral resolution enables us to better characterize and separate similar spectra, statistical analysis of hyperspectral data poses special problems. In this project, we will develop new m ethods for robust analysis of high-dimensional data, in particular for feature selection from the spectral curve, with the focus of landcover classification.