Spectral imaging classification by non linear spectral unmixing algorithm and spectral angle mapper classification

Abstract

Remote sensing is the science that searches for the information by sensing, recording, processing and analyzing the energy emitted or reflected from the surface of Earth. Remote sensing images are captured by aircraft carrying imaging platforms, or by satellite, But due to the high altitudes, the high cost of high-resolution cameras, and the great similarity in the appearance of the scenes we created spectral cubes using spectral imaging technology that depends on producing a group of three-dimensional images with two spatial dimensions (geographic) and a third dimension represents the different wavelengths.Spectral imaging results are a set of pixels that represent a mixture of pure materials and components called (mixed pixel), these components can be distinguished by non linear spectral unmixing automatically. In this paper, the non-linear unmixing algorithm was studied and applied to the spectral cubes resulting from remote sensing images in order to classify the contents of these images, identify their components and calculate the ratio of them, The results of unmixing algorithm were also compared with the results of spectral angle mapper algorithm and shoewed us that spectral unmixing algorithm is more superior because it presents excellent results, for example error percentage for roads is 1.3% , for rocks and water is 2.59%, 0%  respectevely.