The Application of Optimized ICA Method Towards Endmember Extraction

The advent of hyperspectral image technology is a major leap in recent years, it obtains the surface of the earth image contains rich space, radiation and spectral information, Mixed pixels not only effects identification and classification precision of object, but also greatly hinder the development of quantitative remote sensing, so effectively interpret mixed pixels is an important problem for its applications. Based on optimized ICA method a novel hyperspectral unmixing approach is proposed in the paper, which introducing the similarity threshold technique to describe the statistical distribution of the pixels, and determine the criterion of candidate endmembers, A multi-core parallel processing method is proposed to increase its efficiency. The approach is capable of self-adaptation, and can be applied to hyperspectral images with different characteristics. Experimental results on both simulated and real hyperspectral data demonstrated that the proposed approach can provided an effective technique for the blind unmixing and obviously increase the processing efficiency and obtain accurate results at the same time.