Showing papers on "Kernel adaptive filter published in 2018"

Learning Spatial-Aware Regressions for Visual Tracking

Abstract: In this paper, we analyze the spatial information of deep features, and propose two complementary regressions for robust visual tracking. First, we propose a kernelized ridge regression model wherein the kernel value is defined as the weighted sum of similarity scores of all pairs of patches between two samples. We show that this model can be formulated as a neural network and thus can be efficiently solved. Second, we propose a fully convolutional neural network with spatially regularized kernels, through which the filter kernel corresponding to each output channel is forced to focus on a specific region of the target. Distance transform pooling is further exploited to determine the effectiveness of each output channel of the convolution layer. The outputs from the kernelized ridge regression model and the fully convolutional neural network are combined to obtain the ultimate response. Experimental results on two benchmark datasets validate the effectiveness of the proposed method.

Fuzzy-Affine-Model-Based Memory Filter Design of Nonlinear Systems With Time-Varying Delay

Abstract: This paper studies the piecewise-affine memory $\mathscr {H}_{\infty }$ filtering problem for nonlinear systems with time-varying delay in a delay-dependent framework. The nonlinear plant is characterized by a continuous-time Takagi–Sugeno fuzzy-affine model with parametric uncertainties. The purpose is to develop a new approach for filter synthesis procedure with less conservatism. Specifically, by constructing a novel Lyapunov–Krasovskii functional, together with a Wirtinger-based integral inequality, reciprocally convex inequality and S-procedure, an improved criterion is first attained for analyzing the $\mathscr {H}_{\infty }$ performance of the filtering error system, and then via some linearization techniques, the piecewise-affine memory filter synthesis is carried out. It is shown that the existence of desired filter gains can be explicitly determined by the solution of a convex optimization problem. Finally, simulation studies are presented to reveal the effectiveness and less conservatism of the developed approaches. It is anticipated that the proposed scheme can be further extended to the analysis and synthesis of continuous-time fuzzy-affine dynamic systems with integrated communication delays in the networked circumstance.

The Two-State Implicit Filter Recursive Estimation for Mobile Robots

Abstract: This letter deals with recursive filtering for dynamic systems where an explicit process model is not easily devisable. Most Bayesian filters assume the availability of such an explicit process model, and thus may require additional assumptions or fail to properly leverage all available information. In contrast, we propose a filter that employs a purely residual-based modeling of the available information and thus achieves higher modeling flexibility. While this letter is related to the descriptor Kalman filter, it also represents a step toward batch optimization and allows the integration of further techniques, such as robust weighting for outlier rejection. We derive recursive filter equations that exhibit similar computational complexity when compared to their Kalman filter counterpart—the extended information filter. The applicability of the proposed approach is experimentally confirmed on two different real mobile robotic state estimation problems.

Fault location based on travelling wave identification using an adaptive extended Kalman filter

Abstract: The fault location in transmission systems remains a challenging problem, primarily due to the fault location near the substation ends or the weak fault signals. In this study, an adaptive extended Kalman filter (EKF) based on the maximum likelihood (ML) is proposed to estimate the instantaneous amplitudes of the travelling waves. In this method, the EKF algorithm is used to estimate the optimal states (the clean travelling waves) with additive white noise while ML is used to adaptively optimise the error covariance matrices and the initial conditions of the EKF algorithm. Using the proposed method, the singularity points of travelling waves can be detected, and the exact arrival time of the initial wave head at the substations M and N can be easily yielded. Thus the fault distance can be calculated precisely. The effectiveness of exacting mutation feature using the proposed method has been demonstrated by a simulated instantaneous pulse. Also, the proposed method has been tested with different types of faults, such as different fault locations, different fault resistances and different fault inception angles using ATP simulation. The accuracy of fault location using the proposed method has been compared with conventional wavelet transformation scheme.

Novel hybrid estimator based on model reference adaptive system and extended Kalman filter for speed-sensorless induction motor control:

Abstract: This paper presents a novel hybrid estimator consisting of an extended Kalman filter (EKF) and an active power-based model reference adaptive system (AP-MRAS) in order to solve simultaneous estimat...

Graph polynomial filter for signal denoising

Abstract: A technique for denoising signals defined over graphs was recently proposed by Chen et al. (2014). The technique is based on a regularisation framework and denoising is achieved by solving an optimisation problem. Matrix inversion is required and an approximate solution that avoids directly calculating the inverse, by using a graph filter, was proposed by Chen et al . (2014). The technique, however, requires an eigendecomposition and the resulting filter degree is high. In this study, the authors propose a computationally efficient technique that is based on a least squares approximation of the eigenvalues of the inverse. They show that a good approximation can be achieved with a low degree graph polynomial filter without the need for any eigendecomposition. Low degree filters also have the desirable property of vertex localisation (analogous to time localisation). The filter gives denoising results that are very similar to that using the exact solution and can be implemented using distributed processing.

Distributed Adaptive Learning With Multiple Kernels in Diffusion Networks

Abstract: We propose an adaptive scheme for distributed learning of nonlinear functions by a network of nodes. The proposed algorithm consists of a local adaptation stage utilizing multiple kernels with projections onto hyperslabs and a diffusion stage to achieve consensus on the estimates over the whole network. Multiple kernels are incorporated to enhance the approximation of functions with several high- and low-frequency components common in practical scenarios. We provide a thorough convergence analysis of the proposed scheme based on the metric of the Cartesian product of multiple reproducing kernel Hilbert spaces. To this end, we introduce a modified consensus matrix considering this specific metric and prove its equivalence to the ordinary consensus matrix. Besides, the use of hyperslabs enables a significant reduction of the computational demand with only a minor loss in the performance. Numerical evaluations with synthetic and real data are conducted showing the efficacy of the proposed algorithm compared to the state-of-the-art schemes.

Fast High-Dimensional Bilateral and Nonlocal Means Filtering

Abstract: Existing fast algorithms for bilateral and nonlocal means filtering mostly work with grayscale images. They cannot easily be extended to high-dimensional data such as color and hyperspectral images, patch-based data, flow-fields, etc. In this paper, we propose a fast algorithm for high-dimensional bilateral and nonlocal means filtering. Unlike existing approaches, where the focus is on approximating the data (using quantization) or the filter kernel (via analytic expansions), we locally approximate the kernel using weighted and shifted copies of a Gaussian, where the weights and shifts are inferred from the data. The algorithm emerging from the proposed approximation essentially involves clustering and fast convolutions, and is easy to implement. Moreover, a variant of our algorithm comes with a guarantee (bound) on the approximation error, which is not enjoyed by existing algorithms. We present some results for high-dimensional bilateral and nonlocal means filtering to demonstrate the speed and accuracy of our proposal. Moreover, we also show that our algorithm can outperform state-of-the-art fast approximations in terms of accuracy and timing.

Output-Only Damage Detection of Steel Beam Using Moving Average Filter

Abstract: This paper provides a simple and direct output-only baseline-free method to detect damage from the noisy acceleration data by using Moving Average Filter (MAF) MAF is a convolution approach based on a simple filter kernel (rectangular shape) that works as an averaging method to smooth signal and remove incorporated noise In this paper, a method is proposed to employ MAF to smooth acceleration signals obtained from a series of accelerometers and determine the damage location along a steel beam To verify the proposed method, a simply supported beam was modelled through a 3D numerical simulation and an experimental model under a moving vehicle load The response acceleration data was then recorded at a sampling frequency of 500 Hz Finally, damage location was identified by applying the proposed method The results showed that the proposed method can accurately estimate the damage location from the acceleration signal without applying any frequency filtering or baseline correction

Fast 2D Complex Gabor Filter With Kernel Decomposition

Abstract: 2D complex Gabor filtering has found numerous applications in the fields of computer vision and image processing. Especially, in some applications, it is often needed to compute 2D complex Gabor filter bank consisting of filtering outputs at multiple orientations and frequencies. Although several approaches for fast Gabor filtering have been proposed, they focus primarily on reducing the runtime for performing filtering once at specific orientation and frequency. To obtain the Gabor filter bank, the existing methods are repeatedly applied with respect to multiple orientations and frequencies. In this paper, we propose a novel approach that efficiently computes the 2D complex Gabor filter bank by reducing the computational redundancy that arises when performing filtering at multiple orientations and frequencies. The proposed method first decomposes the Gabor kernel to allow a fast convolution with the Gaussian kernel in a separable manner. This enables reducing the runtime of the Gabor filter bank by reusing intermediate results computed at a specific orientation. By extending this idea, we also propose a fast approach for 2D localized sliding discrete Fourier transform that uses the Gaussian kernel in order to lend spatial localization ability as in the Gabor filter. Experimental results demonstrate that the proposed method runs faster than the state-of-the-art methods, while maintaining similar filtering quality.

Implementation Of MobileNet In A CNN Based Digital Integrated Circuit

Abstract: Operations of a combination of first and second original convolutional layers followed by a short path are replaced by operations of a set of three particular convolutional layers. The first contains 2N×N filter kernels formed by placing said N×N filter kernels of the first original convolutional layer in left side and N×N filter kernels of an identity-value convolutional layer in right side. The second contains 2N×2N filter kernels formed by placing the N×N filter kernels of the second original convolutional layer in upper left corner, N×N filter kernels of an identity-value convolutional layer in lower right corner, and N×N filter kernels of two zero-value convolutional layers in either off-diagonal corner. The third contains N×2N of kernels formed by placing N×N filter kernels of a first identity-value convolutional layer and N×N filter kernels of a second identity-value convolutional layer in a vertical stack. Each filter kernel contains 3×3 filter coefficients.

Stochastic Feedback Based Kalman Filter for Nonlinear Continuous-Discrete Systems

Abstract: For continuous-discrete filtering with unpredictable approximation errors, by proposing the novel stochastic feedback scheme, this note elaborates a closed-loop adaptive Kalman filter for nonlinear continuous-discrete systems. In conventional filters, unknown approximation errors might arise due to the integration/discretization and linearization of continuous model, and ruin the optimality of Kalman theory. As the main contribution of this note, the stochastic feedback based covariance adaption scheme does not require the approximation steps; instead, the posteriori sequence is mined as a feedback to adapt the priori error covariance, so that the unpredictable errors and costly calculations can be reduced or controlled in the novel closed-loop filtering structure. The new approach's advantages in computational cost, adaptability, and accuracy have been demonstrated by the numerical simulations.

Multiresolution Volume Filtering in the Tensor Compressed Domain

Abstract: Signal processing and filter operations are important tools for visual data processing and analysis. Due to GPU memory and bandwidth limitations, it is challenging to apply complex filter operators to large-scale volume data interactively. We propose a novel and fast multiscale compression-domain volume filtering approach integrated into an interactive multiresolution volume visualization framework. In our approach, the raw volume data is decomposed offline into a compact hierarchical multiresolution tensor approximation model. We then demonstrate how convolution filter operators can effectively be applied in the compressed tensor approximation domain. To prevent aliasing due to multiresolution filtering, our solution (a) filters accurately at the full spatial volume resolution at a very low cost in the compressed domain, and (b) reconstructs and displays the filtered result at variable level-of-detail. The proposed system is scalable, allowing interactive display and filtering of large volume datasets that may exceed the available GPU memory. The desired filter kernel mask and size can be modified online, producing immediate visual results.

Adaptive Kalman Filtering for Systems Subject to Randomly Delayed and Lost Measurements

Abstract: This paper investigates the problem of state estimation for discrete-time linear systems where the observation data are transmitted from the sensor to the filter subject to random delay and dropout The loss and latency of the measurements are modeled by a group of Bernoulli distributed random variables with uncertain probabilities, which appear in the Kalman filter parameters An adaptation factor, which is defined by comparing the theoretical and practical values of the innovation covariance, is employed to adjust the filter gains during estimation Simulation results are presented to verify the improved performance of the proposed adaptive filter

A Quaternion Kernel Minimum Error Entropy Adaptive Filter

Abstract: In this paper, we develop a kernel adaptive filter for quaternion data based on minimum error entropy cost function. We apply generalized Hamilton-real (GHR) calculus that is applicable to Hilbert space for evaluating the cost function gradient to develop the quaternion kernel minimum error entropy (MEE) algorithm. The MEE algorithm minimizes Renyis quadratic entropy of the error between the filter output and desired response or indirectly maximizing the error information potential. Here, the approach is applied to quaternions for improving performance for biased or non-Gaussian signals compared with the minimum mean square error criterion of the kernel least mean square algorithm. Simulation results are used to verify the performance of the algorithm. Convergence is very fast and is shown to out-perform existing algorithms.

Performance improvement of shunt active power filter by a new wavelet-based low-pass filter: Theory and implementation:

Abstract: In this paper, a wavelet-based low-pass filter (WBLPF) is proposed to improve the direct current (DC) component separation speed. To this end, three different methods are studied to implement discrete wavelet transform, and then they are compared with each other in term of transient response time, accuracy and computational cost. Afterwards, the conventional low-pass filter is replaced by the new WBLPF in the p-q method. This replacement increases the speed of DC component separation, and consequently the performance of the shunt active power filter is improved. To verify and examine the proposed method, a power system is simulated in MATLAB software and a prototype is implemented in the laboratory. The simulation and experimental results confirm the superiority of the proposed method.

The Right Invariant Nonlinear Complementary Filter for Low Cost Attitude and Heading Estimation of Platforms

Abstract: This paper presents a novel filter with low computational demand to address the problem of orientation estimation of a robotic platform. This is conventionally addressed by extended Kalman filtering of measurements from a sensor suit which mainly includes accelerometers, gyroscopes, and a digital compass. Low cost robotic platforms demand simpler and computationally more efficient methods to address this filtering problem. Hence nonlinear observers with constant gains have emerged to assume this role. The nonlinear complementary filter is a popular choice in this domain which does not require covariance matrix propagation and associated computational overhead in its filtering algorithm. However, the gain tuning procedure of the complementary filter is not optimal, where it is often hand picked by trial and error. This process is counter intuitive to system noise based tuning capability offered by a stochastic filter like the Kalman filter. This paper proposes the right invariant formulation of the complementary filter, which preserves Kalman like system noise based gain tuning capability for the filter. The resulting filter exhibits efficient operation in elementary embedded hardware, intuitive system noise based gain tuning capability and accurate attitude estimation. The performance of the filter is validated using numerical simulations and by experimentally implementing the filter on an ARDrone 2.0 micro aerial vehicle platform.

Online Nonlinear Estimation via Iterative $L^2$ -Space Projections: Reproducing Kernel of Subspace

Abstract: We propose a novel online learning paradigm for nonlinear-function estimation tasks based on the iterative projections in the $L^2$ space with probability measure reflecting the stochastic property of input signals. The proposed learning algorithm exploits the reproducing kernel of the so-called dictionary subspace, based on the fact that any finite-dimensional space of functions has a reproducing kernel characterized by the Gram matrix. The $L^2$ -space geometry provides the best decorrelation property in principle. The proposed learning paradigm is significantly different from the conventional kernel-based learning paradigm in two senses: first, the whole space is not a reproducing kernel Hilbert space; and second, the minimum mean squared error estimator gives the best approximation of the desired nonlinear function in the dictionary subspace. It preserves efficiency in computing the inner product as well as in updating the Gram matrix when the dictionary grows. Monotone approximation, asymptotic optimality, and convergence of the proposed algorithm are analyzed based on the variable-metric version of adaptive projected subgradient method. Numerical examples show the efficacy of the proposed algorithm for real data over a variety of methods including the extended Kalman filter and many batch machine-learning methods such as the multilayer perceptron.

CLAss-Specific Subspace Kernel Representations and Adaptive Margin Slack Minimization for Large Scale Classification

Abstract: In kernel-based classification models, given limited computational power and storage capacity, operations over the full kernel matrix becomes prohibitive. In this paper, we propose a new supervised learning framework using kernel models for sequential data processing. The framework is based on two components that both aim at enhancing the classification capability with a subset selection scheme. The first part is a subspace projection technique in the reproducing kernel Hilbert space using a CLAss-specific Subspace Kernel representation for kernel approximation. In the second part, we propose a novel structural risk minimization algorithm called the adaptive margin slack minimization to iteratively improve the classification accuracy by an adaptive data selection. We motivate each part separately, and then integrate them into learning frameworks for large scale data. We propose two such frameworks: the memory efficient sequential processing for sequential data processing and the parallelized sequential processing for distributed computing with sequential data acquisition. We test our methods on several benchmark data sets and compared with the state-of-the-art techniques to verify the validity of the proposed techniques.

Heterogeneous Multiscale Change-Point Inference and its Application to Ion Channel Recordings

Abstract: Ion channel recordings by the patch clamp technique are a major tool to quantify the electrophysiological dynamics of ion channels in the cell membrane, which is for instance important in medicine for the development of new drugs. In this work, we model these recordings as a time series which is equidistantly sampled from the convolution of a piecewise constant signal disturbed by white noise with a lowpass filter. We focus on nonparametric estimation of the underlying signal, but also discuss how to use these estimations to analyze the recordings. Estimating the underlying signal requires to detect multiple change-points in noisy and filtered Gaussian observations. The variance can be constant in time, but also a varying variance is observed in some measurements. Since this change-point regression problem is very difficult, we start with independent Gaussian observations but with heterogeneous noise. Such a model is of its own interest and has further applications for instance in genetics. For this model, we propose the heterogeneous simultaneous multiscale change-point estimator, H-SMUCE. It estimates the piecewise constant function by minimizing the number of change-points over the acceptance region of a multiscale test which locally adapts to changes in the variance. The multiscale test is a combination of local likelihood ratio tests which are properly calibrated by scale dependent critical values in order to keep a global nominal level alpha, even for finite samples. We show that H-SMUCE controls over- and underestimation of the number of change-points at a given probability for finitely many observations. To this end, new deviation bounds for F-type statistics are derived. We also bound the implicitly defined critical values. By combining these bounds, we obtain simultaneous confidence intervals for the change-point locations and a confidence band for the whole signal. Moreover, it allows us to show that H-SMUCE achieves the optimal detection rate and estimates the number of change-points consistently for vanishing signals, even when the number of change-points is unbounded. The only extra assumption we have to suppose is that the length of the constant segments does not vanished too fast. We compare the performance of H-SMUCE with several state of the art methods in simulations and show how it can be computed efficiently by a pruned dynamic program. An R-package is provided. In a second step we combine these multiscale regression techniques with deconvolution to obtain non-parametric estimators for the ion channel recordings. Truncating the filter kernel and pre-estimating the function values on longer constant segments enable us to perform the deconvolution locally which allows fast computation. Simulations and real data applications confirm that the proposed segmentation methods, JULES and JILTAD, estimate the underlying signal very accurately, even when events occur on small temporal scales, where the smoothing effect of the filter hinders estimation by common methods. Moreover, JILTAD shows still good results when the noise is heterogeneous, a situation for which previously no non-parametric estimation method existed. Also these methods are implemented in R. The usage of these methods is demonstrated in a biochemical study against the context of multidrug-resistant bacteria. We showed statistically significant differences for the interaction of the antibiotic ampicillin with the wild type and with the mutant G103K of the outer membrane channel PorB. These results improves the understanding of potential sources for bacterial resistance and might help to develop new drugs against it to alleviate the severe consequences of multidrug-resistant bacteria.

Robust Adaptive Filtering With $q$ -Gaussian Kernel Mean $p$ -Power Error

Abstract: In this letter, a novel information theoretic measure, namely $\boldsymbol {q}$ -Gaussian kernel mean $\boldsymbol {p}$ -power error (QKMPE), is proposed by defining the mean $\boldsymbol {p}$ -power error in the $\boldsymbol {q}$ -Gaussian kernel space, which is a generalization of the kernel mean $\boldsymbol {p}$ -power error measure. Furthermore, a recursive kernel adaptive filter algorithm, named as recursive least $\boldsymbol {q}$ -Gaussian kernel mean $\boldsymbol {p}$ -power, is derived under the least QKMPE criterion for robust learning in noisy environment. This new proposed algorithm reveals superior performance against Gaussian-type noise as well as the non-Gaussian perturbation, especially when the data contain large outliers. Experimental results in the context of Mackey–Glass time series prediction confirm the effectiveness of the proposed algorithm.

Full-network low-dose CT imaging method and apparatus based on convolution residual network

Abstract: The invention discloses a full-network low-dose CT imaging method and a device based on a convolution residual network. Firstly, the method obtains a plurality of sets of corresponding CT original projection data under low-dose and normal-dose. Secondly, a convolution residual network (CNN1) is established in the projection space, which inputs low dose CT projection data and outputs processed datato reduce the noise and artifacts in low dose CT projection data and improve the signal-to-noise ratio. Then, the projection data is reconstructed into the image space by FBP based on Ramp filter kernel, and the image space is reprocessed based on convolution residual network (CNN2) to reduce the artifacts and noise in the low dose data. The invention can effectively reduce artifacts and noises in low-dose CT data, the data quality can meet the requirements of clinical analysis, diagnosis and the like, and improves the image quality of low-dose CT imaging.

Moving target parameter estimation for MIMO radar based on the improved particle filter

Abstract: In this paper, a novel independent partition Rao-Blackwellized particle filter (IPRBPF) is proposed to estimate the moving target parameters in MIMO radar. Firstly, noticing that the likelihood function is a nonlinear function of the nonlinear position parameters, and that the target motion equation is a linear function of linear parameters such as velocity, acceleration and etc. The nonlinear particle filter is proposed to estimate the nonlinear position parameters and the linear Kalman filter is proposed to estimate the linear parameters. Then a new MIMO radar parameter estimation algorithm based on Rao-Blackwellized particle filter is obtained. Furtherly, considering that the computational complexity will increase dramatically with the targets’ state dimension in the case of multiple targets, the independent partition sampling is put forward to improve the performance of our algorithm, then the IPRBPF algorithm is obtained. Compared with the existing methods, the proposed algorithm can achieve the lower computational complexity and the higher accuracy of parameter estimation. Simulation results demonstrate the advantages of the proposed algorithm.

A Hybrid Dictionary Approach for Distributed Kernel Adaptive Filtering in Diffusion Networks

Abstract: We propose a hybrid dictionary approach for distributed kernel-based adaptive learning of a nonlinear function by a network of nodes. The hybrid dictionary incorporates a local part to improve learning of high frequency components in the function within the local domain of each node and a global part to provide a consensus estimate of the function over the whole region of interest. We apply our scheme to the reconstruction of a spatial distribution by a network of mobile nodes. Performance evaluations show that high frequency components are reconstructed accurately by our hybrid dictionary approach while common schemes are not able to recover them completely.

Additive noise removal using a nonlinear hyperbolic PDE-based model

Abstract: A novel second-order partial differential equation (PDE) — based image restoration technique is proposed here. The considered smoothing method is based on a nonlinear hyperbolic differential model combined to a two-dimension filter kernel. The considered PDE model is well-posed and it is solved numerically by developing an explicit iterative finite difference method-based numerical approximation algorithm that is consistent to the combined PDE-based model and is converging fast to its weak solution. Our successful restoration tests and method comparison are also discussed.

Kernel Adaptive Hammerstein Filter

Abstract: To identify Hammerstein systems, a variety of Hammerstein filters have been proposed. However, most of them assume the nonlinear part in Hammerstein systems to be polynomial in the process of modeling, which restricts their applicability in many practical situations. In this paper, a simple kernel adaptive filter (KAF) called kernel least mean square (KLMS) combined with coherence criterion (CC) is used to approximate the nonlinear part of a Hammerstein system, resulting in the kernel adaptive Hammerstein filter (KAHF). The KAHF can identify various Hammerstein systems well without any prior knowledge of nonlinear part. Simulation results confirm the desirable performance of the new method. Index Terms-Hammerstein system identification, kernel adaptive filter, infinite impulse response system