In allusion to the hyperspectral images at higher spatial resolution, it has been demonstrated in this chapter that endmember extraction can be achieved by using an improved target recognition method based on the minimum noise fraction and projection pursuit. The method has a positive effect in reducing noise and removing the correlation between bands. It does not require a priori spectral information. The experiments results show that it is an automatic target recognition algorithm, but the false alarm rate needs to be improved.
展开▼