Singular vector filtering method for mitigation of disturbance enhancement in multichannel active noise control systems

In the design of multichannel active noise control filters, the disturbance enhancement phenomenon will sometimes occur, i.e., the resulting sound is enhanced instead of being reduced in some frequency bands, if the control filter is designed to minimize the power of error signals in other frequency bands or across all frequencies. In previous work, a truncated singular value decomposition method was applied to the system autocorrelation matrix to mitigate the disturbance enhancement. Some small singular values and the associated singular vectors are removed, if they are responsible for unwanted disturbance enhancement in some frequency bands. However, some of these removed singular vectors may still contribute to the noise control performance in other frequency bands; thus, a direct truncation will degrade the noise control performance. In the present work, through an additional filtering process, the set of singular vectors that causes the disturbance enhancement is replaced by a set of new singular vectors whose frequency responses are attenuated in the frequency band where disturbance enhancement occurs, while the frequency responses in other frequency bands are unchanged. Compared with truncation approach, the proposed method can maintain the performance in the noise reduction bands, while mitigating the influence in disturbance enhancement bands.

Sparse Nonnegative Matrix Factorization for Hyperspectral Unmixing Based on Endmember Independence and Spatial Weighted Abundance

Hyperspectral image unmixing is an important task for remote sensing image processing. It aims at decomposing the mixed pixel of the image to identify a set of constituent materials called endmembers and to obtain their proportions named abundances. Recently, number of algorithms based on sparse nonnegative matrix factorization (NMF) have been widely used in hyperspectral unmixing with good performance. However, these sparse NMF algorithms only consider the correlation characteristics of abundance and usually just take the Euclidean structure of data into account, which can make the extracted endmembers become inaccurate. Therefore, with the aim of addressing this problem, we present a sparse NMF algorithm based on endmember independence and spatial weighted abundance in this paper. Firstly, it is assumed that the extracted endmembers should be independent from each other. Thus, by utilizing the autocorrelation matrix of endmembers, the constraint based on endmember independence is to be constructed in the model. In addition, two spatial weights for abundance by neighborhood pixels and correlation coefficient are proposed to make the estimated abundance smoother so as to further explore the underlying structure of hyperspectral data. The proposed algorithm not only considers the relevant characteristics of endmembers and abundances simultaneously, but also makes full use of the spatial-spectral information in the image, achieving a more desired unmixing performance. The experiment results on several data sets further verify the effectiveness of the proposed algorithm.

Optimal preconditioned regularization of least mean squares algorithm for robust online learning1

Despite its low computational cost, and steady state behavior, some well known drawbacks of the least means squares (LMS) algorithm are: slow rate of convergence and unstable behaviour for ill conditioned autocorrelation matrices of input signals. Several modified algorithms have been presented with better convergence speed, however most of these algorithms are expensive in terms of computational cost and time, and sometimes deviate from optimal Wiener solution that results in a biased solution of online estimation problem. In this paper, the inverse Cholesky factor of the input autocorrelation matrix is optimized to pre-whiten input signals and improve the robustness of the LMS algorithm. Furthermore, in order to have an unbiased solution, mean squares deviation (MSD) is minimized by improving convergence in misalignment. This is done by regularizing step-size adaptively in each iteration that helps in developing a highly efficient optimal preconditioned regularized LMS (OPRLMS) algorithm with adaptive step-size. Comparison of OPRLMS algorithm with other LMS based algorithms is given for unknown system identification and noise cancelation from ECG signal, that results in preference of the proposed algorithm over the other variants of LMS algorithm.

Discrete wavelet transform-based RI adaptive algorithm for system identification

In this paper, we propose a new adaptive filtering algorithm for system identification. The algorithm is based on the recursive inverse (RI) adaptive algorithm which suffers from low convergence rates in some applications; i.e., the eigenvalue spread of the autocorrelation matrix is relatively high. The proposed algorithm applies discrete-wavelet transform (DWT) to the input signal which, in turn, helps to overcome the low convergence rate of the RI algorithm with relatively small step-size(s). Different scenarios has been investigated in different noise environments in system identification setting. Experiments demonstrate the advantages of the proposed DWT recursive inverse (DWT-RI) filter in terms of convergence rate and mean-square-error (MSE) compared to the RI, discrete cosine transform LMS (DCTLMS), discrete-wavelet transform LMS (DWT-LMS) and recursive-least-squares (RLS) algorithms under same conditions.

Supervised Distance-Based Feature Selection for Hyperspectral Target Detection

Feature/band selection (FS/BS) for target detection (TD) attempts to select features/bands that increase the discrimination between the target and the image background. Moreover, TD usually suffers from background interference. Therefore, bands that help detectors to effectively suppress the background and magnify the target signal are considered to be more useful. In this regard, three supervised distance-based filter FS methods are proposed in this paper. The first method is based on the TD concept. It uses the image autocorrelation matrix and the target signature in the detection space (DS) for FS. Features that increase the first-norm distance between the target energy and the mean energy of the background in DS are selected as optimal. The other two methods use background modeling via image clustering. The cluster mean spectra, along with the target spectrum, are then transferred into DS. Orthogonal subspace projection distance (OSPD) and first-norm distance (FND) are used as two FS criteria to select optimal features. Two datasets, HyMap RIT and SIM.GA, are used for the experiments. Several measures, i.e., true positives (TPs), false alarms (FAs), target detection accuracy (TDA), total negative score (TNS), and the receiver operating characteristics (ROC) area under the curve (AUC) are employed to evaluate the proposed methods and to investigate the impact of FS on the TD performance. The experimental results show that our proposed FS methods, as compared with five existing FS methods, have improving impacts on common target detectors and help them to yield better results.