A perspective on the analysis of hyperspectral data

A very important technological development in sensor technology is the ability to construct multispectral imaging sensors with very large numbers (≈200) of spectral bands. A significant number of such hyperspectral sensor systems are now in use or under various stages of development in various countries. The existence of such sensors raises the question of how best to analyze the data with so many spectral bands. Though existing conventional multispectral analysis methods will still be useful in the hyperspectral era ahead, it becomes clear that they fall short of the ideal in several senses. For example, conventional methods often prove computationally unreasonable for large numbers of bands, they may not enable the extraction of all the information from the data that might otherwise be possible, and they may prove excessively cumbersome for researchers in their work. In the paper, after a brief introduction, some of the basic characteristics fundamental to the analysis process are outlined. This is followed by an example showing how they may be applied to hyperspectral data analysis