New normalized LMS adaptive filter with a variable regularization factor

A new normalized least mean square(NLMS) adaptive filter is first derived from a cost function, which incorporates the conventional one of the NLMS with a minimum-disturbance(MD)constraint. A variable regularization factor(RF) is then employed to control the contribution made by the MD constraint in the cost function. Analysis results show that the RF can be taken as a combination of the step size and regularization parameter in the conventional NLMS. This implies that these parameters can be jointly controlled by simply tuning the RF as the proposed algorithm does. It also demonstrates that the RF can accelerate the convergence rate of the proposed algorithm and its optimal value can be obtained by minimizing the squared noise-free posteriori error. A method for automatically determining the value of the RF is also presented, which is free of any prior knowledge of the noise. While simulation results verify the analytical ones, it is also illustrated that the performance of the proposed algorithm is superior to the state-of-art ones in both the steady-state misalignment and the convergence rate. A novel algorithm is proposed to solve some problems. Simulation results show the effectiveness of the proposed algorithm.

Low-Rank and Sparse Representation with Adaptive Neighborhood Regularization for Hyperspectral Image Classification

Low-Rank and Sparse Representation(LRSR)method has gained popularity in Hyperspectral Image(HSI)processing.However,existing LRSR models rarely exploited spectral-spatial classification of HSI.In this paper,we proposed a novel Low-Rank and Sparse Representation with Adaptive Neighborhood Regularization(LRSR-ANR)method for HSI classification.In the proposed method,we first represent the hyperspectral data via LRSR since it combines both sparsity and low-rankness to maintain global and local data structures simultaneously.The LRSR is optimized by using a mixed Gauss-Seidel and Jacobian Alternating Direction Method of Multipliers(M-ADMM),which converges faster than ADMM.Then to incorporate the spatial information,an ANR scheme is designed by combining Euclidean and Cosine distance metrics to reduce the mixed pixels within a neighborhood.Lastly,the predicted labels are determined by jointly considering the homogeneous pixels in the classification rule of the minimum reconstruction error.Experimental results based on three popular hyperspectral images demonstrate that the proposed method outperforms other related methods in terms of classification accuracy and generalization performance.

A GAUSSIAN MIXTURE MODEL-BASED REGULARIZATION METHOD IN ADAPTIVE IMAGE RESTORATION

A GMM (Gaussian Mixture Model) based adaptive image restoration is proposed in this paper. The feature vectors of pixels are selected and extracted. Pixels are clustered into smooth,edge or detail texture region according to variance-sum criteria function of the feature vectors. Then pa-rameters of GMM are calculated by using the statistical information of these feature vectors. GMM predicts the regularization parameter for each pixel adaptively. Hopfield Neural Network (Hopfield-NN) is used to optimize the objective function of image restoration,and network weight value matrix is updated by the output of GMM. Since GMM is used,the regularization parameters share properties of different kind of regions. In addition,the regularization parameters are different from pixel to pixel. GMM-based regularization method is consistent with human visual system,and it has strong gener-alization capability. Comparing with non-adaptive and some adaptive image restoration algorithms,experimental results show that the proposed algorithm obtains more preferable restored images.

Regularized least-squares migration of simultaneous-source seismic data with adaptive singular spectrum analysis

Simultaneous-source acquisition has been recog- nized as an economic and efficient acquisition method, but the direct imaging of the simultaneous-source data produces migration artifacts because of the interference of adjacent sources. To overcome this problem, we propose the regularized least-squares reverse time migration method （RLSRTM） using the singular spectrum analysis technique that imposes sparseness constraints on the inverted model. Additionally, the difference spectrum theory of singular values is presented so that RLSRTM can be implemented adaptively to eliminate the migration artifacts. With numerical tests on a fiat layer model and a Marmousi model, we validate the superior imaging quality, efficiency and convergence of RLSRTM compared with LSRTM when dealing with simultaneoussource data, incomplete data and noisy data.

AN IMPROVED SPARSITY ADAPTIVE MATCHING PURSUIT ALGORITHM FOR COMPRESSIVE SENSING BASED ON REGULARIZED BACKTRACKING

Sparsity Adaptive Matching Pursuit (SAMP) algorithm is a widely used reconstruction algorithm for compressive sensing in the case that the sparsity is unknown. In order to match the sparsity more accurately, we presented an improved SAMP algorithm based on Regularized Backtracking (SAMP-RB). By adapting a regularized backtracking step to SAMP algorithm in each iteration stage, the proposed algorithm can flexibly remove the inappropriate atoms. The experimental results show that SAMP-RB reconstruction algorithm greatly improves SAMP algorithm both in reconstruction quality and computational time. It has better reconstruction efficiency than most of the available matching pursuit algorithms.

Classification of Selfish and Regular Nodes Based on Reputation Values in MANET Using Adaptive Decision Boundary

A MANET is a cooperative network in which each node has dual responsibilities of forwarding and routing thus node strength is a major factor because a lesser number of nodes reduces network performance. The existing reputation based methods have limitation due to their stricter punishment strategy because they isolate nodes from network participation having lesser reputation value and thus reduce the total strength of nodes in a network. In this paper we have proposed a mathematical model for the classification of nodes in MANETs using adaptive decision boundary. This model classifies nodes in two classes: selfish and regular node as well as it assigns the grade to individual nodes. The grade is computed by counting how many passes are required to classify a node and it is used to define the punishment strategy as well as enhances the reputation definition of traditional reputation based mechanisms. Our work provides the extent of noncooperation that a network can allow depending on the current strength of nodes for the given scenario and thus includes selfish nodes in network participation with warning messages. We have taken a leader node for reputation calculation and classification which saves energy of other nodes as energy is a major challenge of MANET. The leader node finally sends the warning message to low grade nodes and broadcasts the classification list in the MANET that is considered in the routing activity.

A Self-adaptive Learning Rate Principle for Stacked Denoising Autoencoders

Existing research on image classification mainly used the artificial definition as the pre-training of the original image,which cost a lot of time on adjusting parameters.However,the depth of learning algorithm intends to make the computers automatically choose the most suitable features in the training process.The substantial of deep learning is to train mass data and obtain an accurate classification or prediction without any artificial work by constructing a multi-hidden-layer model.However,current deep learning model has problems of local minimums when choosing a constant learning rate to solve non-convex objective cost function in model training.This paper proposes an algorithm based on the Stacked Denoising Autoencoders(SDA)to solve this problem,and gives a contrast of different layer designs to test the performance.A MNIST database of handwritten digits is used to verify the effectiveness of this model..

LAVRENTIEV'S REGULARIZATION METHOD FOR NONLINEAR ILL-POSED EQUATIONS IN BANACH SPACES

In this paper, we deal with nonlinear ill-posed problems involving m-accretive mappings in Banach spaces. We consider a derivative and inverse free method for the imple- mentation of Lavrentiev regularization method. Using general HSlder type source condition we obtain an optimal order error estimate. Also we consider the adaptive parameter choice strategy proposed by Pereverzev and Schock （2005） for choosing the regularization parameter.

Iterative implementation of the adaptive regularization yields optimality

The adaptive regularization method is first proposed by Ryzhikov et al. for the deconvolution in elimination of multiples. This method is stronger than the Tikhonov regularization in the sense that itis adaptive, i.e. it eliminates the small eigenvalues of theadjoint operator when it is nearly singular. We will show in this paper that the adaptive regularization can be implemented iterately. Some properties of the proposed non-stationary iterated adaptive regularization method are analyzed. The rate of convergence for inexact data is proved. Therefore the iterative implementation of the adaptive regularization can yield optimality.

Convergence and Optimality of Adaptive Regularization for Ill-posed Deconvolution Problems in Infinite Spaces

The adaptive regularization method is first proposed by Ryzhikov et al. in [6] for the deconvolution in elimination of multiples which appear frequently in geoscience and remote sensing. They have done experiments to show that this method is very effective. This method is better than the Tikhonov regularization in the sense that it is adaptive, i.e., it automatically eliminates the small eigenvalues of the operator when the operator is near singular. In this paper, we give theoretical analysis about the adaptive regularization. We introduce an a priori strategy and an a posteriori strategy for choosing the regularization parameter, and prove regularities of the adaptive regularization for both strategies. For the former, we show that the order of the convergence rate can approach O（｜｜n｜｜^4v/4v＋1） for some 0 〈 v 〈 1, while for the latter, the order of the convergence rate can be at most O（｜｜n｜｜^2v/2v＋1） for some 0 〈 v 〈 1.

A fast and adaptive method for complex-valued SAR image denoising based on l_k norm regularization

This paper developed a fast and adaptive method for SAR complex image denoising based on lk norm regularization, as viewed from parameters estimation. We firstly establish the relationship between denoising model and ill-posed inverse problem via convex half-quadratic regularization, and compare the difference between the estimator variance obtained from the iterative formula and biased CramerRao bound, which proves the theoretic flaw of the existent methods of parameter selection. Then, the analytic expression of the model solution as the function with respect to the regularization parameter is obtained. On this basis, we study the method for selecting the regularization parameter through minimizing mean-square error of estimators and obtain the final analytic expression, which resulted in the direct calculation, high processing speed, and adaptability. Finally, the effect of regularization parameter selection on the resolution of point targets is analyzed. The experiment results of simulation and real complex-valued SAR images illustrate the validity of the proposed method.

Bayesian Regularized Quantile Regression Analysis Based on Asymmetric Laplace Distribution

In recent years, variable selection based on penalty likelihood methods has aroused great concern. Based on the Gibbs sampling algorithm of asymmetric Laplace distribution, this paper considers the quantile regression with adaptive Lasso and Lasso penalty from a Bayesian point of view. Under the non-Bayesian and Bayesian framework, several regularization quantile regression methods are systematically compared for error terms with different distributions and heteroscedasticity. Under the error term of asymmetric Laplace distribution, statistical simulation results show that the Bayesian regularized quantile regression is superior to other distributions in all quantiles. And based on the asymmetric Laplace distribution, the Bayesian regularized quantile regression approach performs better than the non-Bayesian approach in parameter estimation and prediction. Through real data analyses, we also confirm the above conclusions.

基于自适应整形正则化的AVAz反演

高角度裂缝的存在导致地下介质呈现方位各向异性,具有水平对称轴的横向各向同性介质中反射系数随方位角的变化可以近似表示为傅里叶级数的形式.傅里叶系数的大小取决于背景介质和裂缝参数,傅里叶系数的相位角取决于裂缝对称轴方向.常规傅里叶级数分析利用逐个时间采样点的方位地震数据求和计算傅里叶系数,计算结果易受噪声影响.为了提高傅里叶系数估计的抗噪性和稳定性,将常规傅里叶级数分析与AVAz(振幅随方位角变化)反演方法相结合,提出一种基于整形正则化的AVAz二阶傅里叶系数反演方法.常规整形正则化约束中的整形算子形态固定,提出一种基于方位各向异性强度的自适应整形算子,利用基于常规傅里叶级数分析的二阶傅里叶系数初步估计结果计算阻尼因子,结合一阶差分矩阵构建自适应整形算子,达到在反演中自适应地调整整形算子形态的目的.理论测试表明,基于自适应整形正则化的AVAz反演方法具有较强的抗噪性,能够有效提高二阶傅里叶系数反演的稳定性.实际叠前地震数据应用结果显示裂缝预测结果与测井资料相吻合,证明了基于自适应整形正则化AVAz反演进行裂缝预测的有效性.

基于扩散模型和爬坡趋势分类的风电功率自适应区间预测

扩散模型基于马尔可夫链的概率性质,能够定量描述风电的随机性和不确定性。然而,传统基于扩散模型的时序预测方法以当前输入前一段样本的均值作为基准进行特征缩放,导致预测区间在高峰时段过大、低谷时段过小。因此,提出一种基于扩散模型和爬坡趋势分类的风电功率自适应区间预测方法。首先,利用基于扩散模型的区间预测框架获取初始预测区间。然后,将风电波动过程划分为6种模式,对不同模式下的预测区间采取自适应规整策略,进而获得初始改进区间。接着,针对高出力模式中非爬坡时段的区间带宽不匹配问题,建立爬坡趋势分类评估模型,并结合所属出力模式进行区间修正,获得最终的区间预测结果。最后,实验结果表明所提方法的区间预测效果更优。

基于自适应动态粒子群优化的RAK-SVD方法

K均值奇异值分解(K-SVD)算法是一种行之有效的地震资料去噪方法,但由于其稀疏分解存在不确定性,需要引入正则项对其改进。为此,在常规粒子群算法的基础上,提出了一种自适应动态粒子群算法优化正则化参数的正则化近似K-SVD(RAK-SVD)去噪方法。首先通过修改字典原子和相关参数,解决了由于常规粒子群算法的惯性参数固定不变,导致后期搜索效率下降的问题;其次将正则化系数引入近似K-SVD(AK-SVD)方法,明显提升了去噪效果;最后利用自适应动态粒子群算法自动优选AK-SVD方法中的正则化参数,提高了稀疏分解的确定性,在对强反射信号进行去噪的同时加强了对弱信号的保护。模型测试和实际应用均表明,该方法有利于弱信号的提取和识别,不仅能够显著改善弱地震信号的去噪效果,还提升了计算效率。该方法具有一定的实际应用价值。

动态时空异常感知的相关滤波目标跟踪算法

针对背景感知算法未与目标的时空域特性建立联系,以及无法准确处理遮挡、形变等异常跟踪情况的问题,提出了能够动态感知时空异常的目标跟踪算法。在相关滤波器训练过程中引入动态空间正则项,使其与样本的时空域特性建立联系;结合响应图的峰值唯一性和锐利信息,提出异常感知方法;利用历史滤波器具有不同置信度的特点以及目标在时域中的连续性,通过异常感知方法自适应选择高置信度的历史滤波器作为时间正则化的参考模板,降低滤波器退化的风险。在OTB50、OTB100和TC128测试基准上进行仿真实验,该算法能够适应外观变化、画面杂乱等复杂条件下的跟踪任务,具有较强的鲁棒性和实用性。

UPRE方法在图像恢复正则化参数自适应选择中的应用

大多数图像恢复都是不适定问题,需要利用正则化方法将病态方程转化为适定方程。正则化参数主要用于调节图像保真度与图像光滑度之间的平衡,与图像恢复质量的好坏有着密切关系。无偏预计风险估计(UPRE)方法可以自适应选择最佳正则化参数,基于数学理论分析和图像恢复实验,验证了该方法在图像恢复中的可行性和有效性。

基于粗到细的多尺度单幅图像去雾方法

为了解决现有图像去雾算法易出现细节纹理丢失、颜色失真或对非均匀浓雾处理不彻底的问题,提出一种基于粗到细的多尺度单幅图像去雾方法。首先,主干网络使用残差特征注意力模块对有雾图像进行特征提取;其次,将不同尺度的输入图像进行卷积预处理,通过多尺度特征融合模块将预处理的浅层特征与主干网络融合;再次,将不同粒度的非对称特征进行有效融合;最后,将浅层信息与深层信息自适应混合输出,通过对比正则损失构建正负样本信息,使得去雾图像更接近无雾图像。实验结果表明,与已有代表性的去雾方法相比,提出的方法能对合成数据集与真实数据集进行有效去雾,在细节保留与色彩还原上优于对比方法。

三维大地电磁测深阶段式自适应正则化反演

如何合理地确定正则化因子是地球物理正则化反演领域的研究热点。阶段式自适应算法可以充分发挥模型稳定器的作用,提高反演结果的稳定性,但是该算法仅在一维、二维大地电磁测深(Magnetotelluric,MT)反演中得以实现。目前,三维MT反演正在快速发展,基于此,本文将阶段式自适应正则化算法引入三维MT正则化反演,按照“阶段”而不是“迭代次数”自适应地调整正则化因子的取值,进而观察反演结果的变化。本文设计单块体和双块体模型试验,并特意设置了较大的迭代次数,进而观察反演结果随反演进程的变化情况。模型试验表明:阶段式自适应算法是适用于三维MT正则化反演的,该算法在反演的后期可以更好地保持解的稳定,故此,从解的稳定性这个角度去考量正则化因子的选择是一种值得探索的方向。

各向异性的L_(0)正则化图像平滑方法

现有的图像平滑方法缺乏灵活性,会导致边缘不清晰、结构缺失和过度锐化等问题。文中提出一种新的自适应加权矩阵的正则化方法,主要应用于图像平滑,并且可以扩展到其他应用。提出的模型设计了一个新的正则化项,基于梯度算子▽和自适应加权矩阵T组合为L_(0)范数正则化项,使得模型具有各向异性。通过为不同梯度方向赋予不同的权重,以此来刻画平滑图像的局部结构,更好地展现局部特征,防止过度平滑。由于所提出的模型是非光滑且非凸的,在求解上比较复杂,因此采用ADMM算法对模型进行求解。把目标函数分解成几个易求解的子问题,分别对每个子问题求解,最终得到模型的最优解。主客观实验表明,提出的模型在视觉效果以及数值方面都有明显的提高。