The Research Of Endmembers Extraction From Hyperspectral Remote Sensing Image Based On Locally Linear Embedding

Spectral remote sensing technology has been widely used in civil and military fields. And its band width can reach nanometer order. Hyperspectral remote sensing can obtain more precise data, more bands, and provide more detailed and rich feature information for the study of remote sensing technology.However, due to the limited spatial resolution of spectrometer, in hyperspectral image, the mixed pixels are existed widely. Therefore, the mixed pixel cannot be used in analysis directly. It is necessary to apply the process of hyperspectral image unmixing. And endmember extraction is critical to the spectral unmixing. Endmember extraction is to extract the “pure” spectrum which called endmember from the mixed pixel. How to extract endmember accurately and then carry out the unmixing effectively has become a hot issue in the study of spectrum processing algorithm.This paper mainly research on endmember extraction algorithms for hyperspectral image. We focus on the N-FINDR algorithm, which is the most widely used endmember extraction algorithm. And the improved N-FINDR algorithm based on the nonlinear dimensionality reduction is proposed in this paper. The purpose is to solve the nonlinear spectral mixture problem for hyperspectral image. We use nonlinear dimensionality reduction by the improved locally linear embedding algorithm based on robust spatial information(RSLLE). RSLLE is applied to achieve the data after the nonlinear dimensionality reduction. Then N-FINDR algorithm is used to extract endmember. Thus, the improved N-FINDR algorithm can be obtained, which called RSLLE-N-FINDR in this paper. Finally, two experiments: synthetic data experiment and real hyperspectral images experiment are performed. And the results show that the proposed RSLLE-N-FINDR method achieves great precision of endmember extraction.