Discrete Artificial Bee Colony Algorithm For Endmember Extraction Of Hyperspectral Remote Sensing Image

Spectral unmixing is an important topic in hyperspectral image exploitation. It comprises of extraction of a set of pure spectral signatures and their corresponding fractional abundance in each pixel. The key step of unmixing is endmember extraction. In the past years, many algorithms for endmember extraction have been developed based on swarm intelligence techniques, such as, DPSO-EE, ACO-EE, etc. The basic model of spectral unmixing is that the geometry description follows a Linear Mixture Model(LMM). According to our experience, we have found out there are some inevitable disadvantages in the umixing process. First and foremost, the error accumulates due to the fact that there is no information from a feedback mechanism. Also some maladjustments may appear when dealing with images having big errors or strong noises for simple evaluation of endmembers.In this thesis, we present a new optimal model for endmember extraction, namely, a new swarm intelligent method based on discrete artificial bee colony algorithm for endmember extraction(DABC-EE). This method could resolve the aforementioned issues and lead to good estimation of the endmembers,meanwhile locate positions of endmembers. This thesis also summarizes thirteen popular and new developing algorithms, including PPI、IEA、AVMAX、MVES、SPA、N-FINDR、VCA、SPICE、PCOMMEND、MVSA、MVC-NMF、DPSO and ACO-EE, which are classified into different four way—pure pixel based, minimum volume based, statistical method based, and swarm intelligence techniques based. A comprehensive comparison and analysis concludes the advantages and disadvantages of each of the thirteen algorithms.This thesis are made of the five chapters. It introduce research signification and progress of endmember extraction for hyperspectral image unmixing, introducing the causes of mixed pixel, sketching the technological process of unmixing in the first chapter and second chapter. The fourth chapter of the thesis detail with the discrete artificial bee colony algorithm for endmember extraction. In the fifth chapter, a comprehensive comparison and analysis concludes the advantages and disadvantages of every algorithms, and further to support the advantage of proposed method. In future, embracing these algorithms to process relational study, the research would offer theoretical support and consult.