**5. Conclusions**

Band selection has been always a challenge in the processing of high-dimensional hyperspectral data. In this paper, a novel method was presented to select a subset of bands that led especially to improving the results of spectral unmixing. The proposed method—named ISI-PS—integrates two measures of band selection. Firstly, it is aimed at managing the spectral variability. To do so, the bands were prioritized in a way so as to have the least inter-class variability while at the same time achieving the highest possible between-class separation. On the other hand, the second phase takes into account the bands' dependency and makes an effort to detect and remove highly correlated bands. This phase was performed in the Prototype Space, which was formed by image endmembers. In the Prototype Space—in which the bands were treated as the space points—bands' dependencies were examined via their inter-angles.

As mentioned above, the second phase of the proposed method required the knowledge of image endmembers, which is itself a challenge in hyperspectral image processing. In this paper, as with the other contribution, an unsupervised automatic technique was proposed that can effectively extract the endmembers from the image itself and that needed no more input knowledge.

The proposed method was examined and validated on a variety of simulated and real datasets. To do so, the selected bands were used in the spectral unmixing, and the RMSE of the obtained fractional abundances was considered as the accuracy measure. The obtained results were all compatible with the in-situ observations and confirmed the effectiveness of the proposed method. In addition, the performance of the proposed method was compared with the SZU and the MTD algorithms, which proved the superiority of the proposed method.

**Author Contributions:** All of the authors listed contributed equally to the work presented in this paper.

**Conflicts of Interest:** The authors declare no conflict of interest.
