Comparison of Unsupervised Band Selection Methods for Hyperspectral Imaging
IbPRIA '07 Proceedings of the 3rd Iberian conference on Pattern Recognition and Image Analysis, Part I
| Hi-index | 0.00 |
Selecting the most relevant bands from a hyperspectral image would considerably reduce the amount of data without practically losing relevant information. In addition, if some physical and signal criteria of this selection are taken into account, the obtained results grouping consecutive bands would be useful to design new filters for hyperspectral cameras in order to improve the efficiency of these devices. Starting from certain number of pre-selected bands, intervals of spectrally adjacent instances to these initial bands are considered for calculating new broader bands. Results will show how a weighted average on these intervals can keep, or even improve, the performance respecting to a narrower selection, avoiding, at the same time, common drawbacks from the narrow-band acquisition devices.