Rank estimation of parafac reducing both signal-dependent and signal-independent noise in hyperspectral image for target detection

One of the most important applications of hyperspectral im- age (HSI) is target detection which aims to detect the pres- ence of a signal of interest embedded in noise. This paper shows that both the signal-dependent (SD) and the signal- independent (SI) noise can be removed by applying a multi- linear algebra decomposition, namely the parallel factor anal- ysis (PARAFAC) decomposition, but the rank estimation of PARAFAC decomposition is time consuming. By analyzing the relationship between the rank value of PARAFAC decom- position and the target detection results, the initial value of the iteration to estimate the optimal rank can be set appro- priately instead of the cycle from 1 to start. The simulaitons show that the computing time can be reduced significantly by using this initialization strategy without affecting the target detection results.