Advantages and limitations of segmentation for point target detection in hyperspectral imagery

摘要

Segmentation appears to be an attractive preprocessing state when performing point target detection in hyperspectral data. Nevertheless, the literature contains examples of both successful and unsuccessful segmentation. Using simulations and several new analytical tools, we propose to derive guidelines when segmentation would be useful and when it would be superfluous. A real dataset with parameters when segmentation is worthwhile and when not is given.