Search

To locate multiple sources through time-difference-of-arrival (TDOA) measurements, existing algorithms generally require the matching relationship between measurements and the corresponding sources. In this paper, we propose a new Bayesian learning method for cases where the matching relationship is not given and off-grid error is considered. To achieve this, first we propose a new basis generator, which casts the localization problem within the Bayesian learning scheme. Then, we modify the existing sparse Bayesian inference (SBI) approaches and explore the priors on fingerprinting weights, resulting in two intermediate algorithms. On these foundations, a subspace-based robust SBI (SRSBI) algorithm is proposed as the core of this paper. SRSBI is highlighted by its ability to work free from iteration when estimating off-grid targets. What' more, SRSBI offers considerable robustness against initial guesses of hyper-parameters. Numerical simulations demonstrate the superiority of SRSBI in terms of accuracy, robustness and speed, compared to the other reported ones. [ABSTRACT FROM AUTHOR]

Abstract: In this paper, the problem of despeckling SAR images when the input data is either an intensity or an amplitude signal is revisited. State-of-the-art despeckling methods based on Bayesian estimators in the wavelet domain, recently proposed in the literature, are taken into consideration. First, how these methods proposed for one format (e.g., intensity) can be adapted to the other format (e.g., amplitude) is investigated. Second, the performance of such algorithms in both cases is analyzed. Experimental results carried out on simulated speckled images and on true SAR data are presented and discussed in order to assess the best strategy. From these results, it can be observed that filtering in the amplitude domain yields better performances in terms of objective quality indexes, such as preservation of structural details, as well as in terms of visual inspection of the filtered SAR data. [Copyright &y& Elsevier]