Spectral unmixing based fusion algorithm for hyperspectral and multi-spectral images

摘要

The intention for fusion of low-spatial-resolution hyperspectral image (HSI) and high-spatial-resolution multi-spectral image (MSI) is to obtain a high-resolution image in both spectral and spatial domains, which is more suitable for object identification and precise positioning. A novel fusion algorithm for HSI and MSI is proposed in this paper to obtain a better spectral preserving performance. The method commences by endmembers extraction to extract the pure object spectra in HSI, then the abundances of these objects obtained by spectral unmixing are used to classify the pixels in MSI and determine which object these pixels belong to. Once the endmembers extraction and spectral unmixing results are obtained, a fused HSI with almost no distortion in spectra will be reconstructed by assigning the pure object spectra to pixels of MSI. The experimental results demonstrate that the proposed algorithm provides an effective way for fusion of HSI and MSI and avoids the spectral distortion to the most extent.