Multiple-model MKS with improved learning/prior modeling

Estimation of topography and fusion of data acquired over multiple resolutions have been extensively studied over the years. The standard MKS (multiscale Kalman smoother) algorithm embedded with a single stochastic model parameterized using power spectra matching methods have been found to give suboptimal performance in estimating nonstationary topographic variations. In this work, multiple models are regulated by a mixture-of-experts (MOE) network to adaptively fuse the estimates. A fractal based approach was employed to segment the data and parameterize the multiple models for better performance.