A higher order statistical approach to spectral unmixing of remote sensing imagery

In this paper, a novel approach for unsupervised spectral unmixing in remote sensing imagery is presented. This approach is derived from independent component analysis (ICA). First, we present the limitations of Gaussian mixture model (GMM) and ICA for spectral unmixing. To overcome these limitations we have developed an approach that employs the ICA model to characterize the data generation process and have proposed an ICA mixture model (ICAMM) based approach for unsupervised spectral unmixing. This approach estimates the endmember probability density function by modeling it with a nonGaussian probability distribution. Thus, the ability to model higher order statistical properties of remote sensing imagery increases the practical applicability of ICAMM for spectral unmixing. The results from our experimental study have demonstrated the efficacy of the proposed algorithm for unsupervised spectral unmixing.