The Research And Application Of The Effective Endmember Set Based On Bilinear Unmixing Model

With the characteristic of abundant band information, high resolution and map one, hyperspectral remote sensing image is widely used in geological exploration, marine monitoring, military reconnaissance and many other fields. But affected by the limitation of remote sensor spatial resolution and the complex surface features in imaging process, there are mixed pixels with a variety of surface features in hyperspectral images. It hinders the depth development of quantificational remote sensing. Therefore, decompose mixed pixels has far-reaching meaning on the development of hyperspectral remote sensing technology.The spectral unmixing technology is to use the pure pixels (called "endmember") to decompose the mixed pixels based on mixing model and unmixing algorithm which can obtain the abundance of different endmember ingredients. The linear mixing model does not account for the interaction of multiple scattering of photons leading to the low accuracy of abundance. This paper analyzes the physical significance, imaging mechanism and application areas among different mixing models, improves the bilinear mixing model to enhance unmixing accuracy. Researches include the following aspects:(1) Although the bilinear unmixing model takes into account the role of scattering of photons, more endmember types in unmixing process, high computational complexity and affecting unmixing accuracy. In fact, different mixed pixels contain different types of endmembers, and the majority of pixels contain fewer categories. The paper puts forward effective endmembers based bilinear mixing model (EEBBMM). Only select the effective endmembers to unmix. And test the spectral characteristics between mixed pixels and endmembers in simulate data, such as spectral angle, Euclidean distance, correlation coefficient and shape spectral angle. Spectral angle and Euclidean distance with highly complementary are selected to optimize the spectral characteristics. Combined with the characteristics of the reconstruction error to determine the effective endmembers set. The experimental results of simulation images illustrate the validity of the model.(2) Different mixing models are applied to the actual oil spill images. The result by gasoline spectral data measured in laboratory and three different types of oil spill image not only explain bilinear mixing model than linear mixing model is more suitable for the layered structure of oil and water, but also illustrate the feasibility of the application of EEBBMM in the actual. Meanwhile, compared with the other endmember set selection algorithm, it further illustrates the validity of proposed effective endmember set selection algorithm.