Intrinsic Bayesian model for high-dimensional unsupervised reduction

摘要: This paper proposes a novel algorithm for high-dimensional unsupervised reduction from intrinsic Bayesian model. The proposed is to assume that the pixel reflectance results nonlinear combinations of pure component spectra contaminated by additive noise. constraints are naturally expressed in literature using appropriate abundance prior distributions. posterior distributions unknown model parameters then derived. consists inductive cognition part and hierarchical part. has several advantages over traditional distance based on algorithms. cognitive used decide which dimensions advantageous output recommended hyperspectral image. can be interpreted as fast inference method We describe procedures learning hyperparameters, computing distribution, extensions Experimental data demonstrate robust useful properties algorithm.