Data fusion of multispectral remote sensing measurements using wavelet transform

Abstract

This thesis focuses on fusion of multispectral data available from remote sensing instruments. The aim is to develop fast and memory efficient algorithms that may be used for real-time implementation aboard satellites. Multiple channel data from the SSM/I instrument are used for experiments. Starting with a Bayesian estimation formulation of the data fusion problem, an attempt is made to take advantage of the sparseness resulting from wavelet transforms to optimize computational efficiency. After generating the necessary statistical models for the data to be estimated, a preconditioning whitening filter, which simplifies the choice of the required wavelet transform, is developed. The significant gains obtained by a compact representation in wavelet basis are shown. An input grid transformation leading to channel filters is then used to construct a real-time implementation of the optimal estimator. Simulated results of such a system are then used to demonstrate the achieved improvement in field resolution.

In conclusion, a direction for future work is laid out for improving the estimation optimality over non-stationarity by adaptive techniques and extension to future instruments.