Evaluation the performance of optimum endmembers extraction method for hyperspectral image change detection

Interests in hyperspectral images (HSIs) has increased significantly over the past decade, essentially due to high spectral resolution, which enables precise material extraction and identification. Hyperspectral change detection (HSCD) is used in many applications ranging from environmental monitoring to city planning and military surveillance. Change Detection (CD) by unmixing has the potential to provide subpixel information. Hyperspectral unmixing is a powerful tools for analysing hyperspectral data. Endmember extraction is a vital step in spectral unmixing of HSIs. Endmembers refer to the pure materials’ spectra in HSIs, and endmember extraction is a process of finding the spectra of all the endmembers. Endmembers minimize the false changes and decrease the Hughes phenomenon effect. This work presents a strategy for multiple CD, which provides a better figure of merit in terms of optimum endmember extraction versus computational complexity. In this paper, we evaluate a hyperspectral unmixing algorithm in multiple CD by extracting endmembers optimally by similarity assessment and finally represent change maps. The experiment for real hyperspectral data demonstrated that both the CD accuracy and the computational performance promoted from proposed strategy. The results demonstrate achieving a higher accuracy using Minimum Distance (MiD) classification method rather than K-Means (KM) classification method for multiple CD.

Keywords:

Hyperspectral , Endmember , Change Detection , FCLS , Abundance , SAM