A Non-Linear Dimensionality Reduction Approach for Unmixing Hyper Spectral Data

This paper offers a non-linear dimensionality discount approach to unmixing hyperspectral records. Unmixing, that is, decomposing hyperspectral records into a hard and fast of element spectra and their respective fractions, is critical in quantifying resources in far-flung sensing. Nevertheless, this has proved challenging because of problems that include high dimensionality and noise in the information. The technique provided uses non-linear manifold mastering algorithms to generate low-dimensional representations of the information. Unmixing is then achieved on those low-dimensional representations, and similarly, refinement is applied using linear strategies. Experiments were carried out on simulated datasets to assess the technique's effectiveness, with top outcomes being received.