Improvement of target detection based on tensorial modelling

In this paper a multichannel and multicomponent restoration scheme is introduced for hyperspectral images (HSI) with the aim of improving target detection. This noise reduction (NR) method takes advantage of the whole data along all dimension simultaneously by defining data as a tensor. The aim of this paper is to prove the improvement in considering the cross-dependency of spatial and spectral information. Using jointly spatial and spectral processing enables better spectral signature restoration and consequently increase the target discrimination. Defining a tensor model, our method is based on tensor decomposition without any dimensional splitting during the processing. The optimization criterion used is the minimization of the mean square error between the estimated and the desired signals. This minimization leads to some estimated n-mode filters for each dimension, which can be considered as the extension of the well-known Wiener filter in a particular mode (such that dimension). In order to take into account the mode cross-dependency, an Alternating Least Square (ALS) algorithm is proposed to jointly determine the n-mode Wiener filter. Comparative studies with the classical bidimensional filtering methods show that our algorithm presents better performances by improving the detection probability.