Enhancing pure-pixel identification performance via preconditioning

摘要

In the scenario where pure pixels exist, pure-pixel search is well known to be a simple and effective approach for hyperspectral unmixing. However, in the noisy case, the performance of a pure-pixel search algorithm usually depends on the conditioning of the endmember matrix. This paper describes several data preconditioning methods for mitigating the aforementioned issue, including a newly proposed heuristic approach. The development is based on a pure-pixel search algorithm called successive projection algorithm (SPA). Simulations based on synthetic data sets show that preconditioning makes SPA much more robust against noise.