The ultra-spectral image set aims to extract the spectra of pure materials from the same scene (the final member bars or orange color), as well as the average amount per pixel of the image. Most algorithms rely on traditional end-organ exploration techniques that often did not work in difficult scenarios. In this work, this problem will be addressed along with the variation of the material by considering that the final member is a direction in the surrounding space which in this case is a one-point substitute. Under this work, we will propose a new algorithm through which we generate a strong reference spectrum. We see the potential of this proposed algorithm which we will apply using real spectra to a set of synthetic data with variability within the class, and the landscape will be the image used.