快速收敛最小方差无畸变响应算法研究及应用

常规最小方差无畸变响应(MVDR)自适应波束形成是一种高分辨窄带波束形成器,它是利用实际声场的窄带互谱密度矩阵(CSDM)估计出自适应波束形成权向量。在实际应用中,MVDR算法需要较长的观测时间估计协方差矩阵,不利于对高速运动目标进行定位;对于宽带目标信号,MVDR算法需要对每一个CSDM进行求逆运算,计算量较大;在相干源条件下,目标信号之间会发生"对消"现象,MVDR算法性能急剧恶化。本文提出了基于子带子阵处理的快速收敛MVDR自适应波束形成方法。首先将全频带划分成一组子带,将接收线阵划分成一组子阵,然后对每一子带计算降维的驾驶协方差矩阵(STCM),从而得到快速收敛MVDR自适应波束形成的权值和空间谱估计结果。同时采用双向空间平滑方法对相干源进行MVDR空间谱估计。仿真和海试数据处理结果表明该算法在保证高分辨力的同时,具有瞬时收敛的性能,双向空间平滑技术具有良好的解相干性能。      

A method for visualization of invariant sets of dynamical systems based on the ergodic partition

We provide an algorithm for visualization of invariant sets of dynamical systems with a smooth invariant measure. The algorithm is based on a constructive proof of the ergodic partition theorem for automorphisms of compact metric spaces. The ergodic partition of a compact metric space A, under the dynamics of a continuous automorphism T, is shown to be the product of measurable partitions of the space induced by the time averages of a set of functions on A. The numerical algorithm consists of computing the time averages of a chosen set of functions and partitioning the phase space into their level sets. The method is applied to the three-dimensional ABC map for which the dynamics was visualized by other methods in Feingold et al. [J. Stat. Phys. 50, 529 (1988)]. (c) 1999 American Institute of Physics.     

SVM参数优化方法分析与决策

针对支持向量机应用过程中的参数选择问题,从UCI数据库选择样本集,分别采用传统的网格法、智能优化算法中的粒子群法及遗传算法实现核函数参数寻优过程,将所得最佳参数应用到样本测试中。在深入分析优化过程中各参数关系、参数对支持向量机性能的影响以及传统与智能优化算法的优劣后,得出了核函数优化策略。即先使用智能优化算法初步确定最优解范围,再结合网格法进行高精度寻优。实验数据验证了参数优化策略的有效性,为扩大支持向量机泛化率、提高应用性做了铺垫。     

基于DFT插值的线性约束最小方差宽带自适应阵列

本文提出一种具有频率不变波束图的线性约束最小方差宽带自适应算法。首先给出了具有频率不变波束图的连续线阵的灵敏度函数与离散线列阵加权系数之间的关系，然后给出了使用DFT插值法求解各子带阵列权系数的方法，最后将DFT插值法应用于线性约束最小方差宽带自适应阵列。理论分析及仿真结果表明，该算法可以在实现最小方差波束形成的同时保持波束图基本不随频率变化，且该方法可以降低宽带自适应阵列的运算量。     

基于RBF网络的断路器电流自适应保护研究

针对传统断路器电流保护方法存在受系统运行方式影响、整定困难、智能化低等问题,本文提出了基于RBF的断路器电流自适应保护算法,并给出了算法的模型。该算法融合了RBF神经网络的故障检测和电流自适应保护。首先通过RBF网络检测负载线路的电流故障,然后用电流自适应算法进行保护。在对神经网络进行训练时,利用PSO算法对RBF神经网络的参数进行优化以此来提高网络的泛化能力和学习能力;然后采用优化后的PSO-RBF神经网络对电流故障进行诊断。实验表明,该算法较大地提高了断路器智能化管理水平。     

Optimizing neural network architectures using generalization error estimators

This paper addresses the optimization of neural network architectures. It is suggested to optimize the architecture by selecting the model with minimal estimated averaged generalization error. We consider a least-squares (LS) criterion for estimating neural network models, i.e., the associated model weights are estimated by minimizing the LS criterion. The quality of a particular estimated model is measured by the average generalization error. This is defined as the expected squared prediction error on a novel input-output sample averaged over all possible training sets. An essential part of the suggested architecture optimization scheme is to calculate an estimate of the average generalization error. We suggest using the GEN-estimator [9, 10] which allows for dealing with nonlinear, incomplete models, i.e., models which are not capable of modeling the underlying nonlinear relationship perfectly. In most neural network applications, it is impossible to suggest a perfect model, and consequently the ability to handle incomplete models is urgent. A concise derivation of the GEN-estimator is provided, and its qualities are demonstrated by comparative numerical studies.The Computational Neural Network Center, Electronics Institute, The Technical University of Denmark, Building 349. Translated from Izvestiya Vysshikh Uchebnykh Zavedenii, Radiofizika, Vol. 37, No. 9, pp. 1131–1147, September, 1994.     

Multiscale top-hat selection transform based infrared and visual image fusion with emphasis on extracting regions of interest

To effectively combine regions of interest in original infrared and visual images, an adaptively weighted infrared and visual image fusion algorithm is developed based on the multiscale top-hat selection transform. First, the multiscale top-hat selection transform using multiscale structuring elements with increasing sizes is discussed. Second, the image regions of the original infrared and visual images at each scale are extracted by using the multiscale top-hat selection transform. Third, the final fusion regions are constructed from the extracted multiscale image regions. Finally, the final fusion regions are combined into a base image calculated from the original images to form the final fusion result. The combination of the final fusion regions uses the adaptive weight strategy, and the weights are adaptively obtained based on the importance of the extracted features. In the paper, we compare seven image fusion methods: wavelet pyramid algorithm (WP), shift invariant discrete wavelet transform algorithm (SIDWT), Laplacian pyramid algorithm (LP), morphological pyramid algorithm (MP), multiscale morphology based algorithm (MSM), center-surround top-hat transform based algorithm (CSTHT), and the proposed multiscale top-hat selection transform based algorithm. These seven methods are compared over five different publicly available image sets using three metrics of spatial frequency, mean gradient, and Q. The results show that the proposed algorithm is effective and may be useful for the applications related to the infrared and visual image fusion.     

On training feedforward neural networks

A new algorithm that mapsn-dimensional binary vectors intom-dimensional binary vectors using 3-layered feedforward neural networks is described. The algorithm is based on a representation of the mapping in terms of the corners of then-dimensional signal cube. The weights to the hidden layer are found by a corner classification algorithm and the weights to the output layer are all equal to 1. Two corner classification algorithms are described. The first one is based on the perceptron algorithm and it performs generalization. The computing power of this algorithm may be gauged from the example that the exclusive-Or problem that requires several thousand iterative steps using the backpropagation algorithm was solved in 8 steps. Another corner classification algorithm presented in this paper does not require any computations to find the weights. However, in its basic form it does not perform generalization.     

Generalization in large committee machines

We study generalization in large committee machines. For a model with nonoverlapping receptive fields a full replica calculation yields results qualitatively similar to those for single-layer machines. For a fully connected architecture, within the annealed approximation we find a transition from a symmetric state to one with specialized hidden units, accompanied by a discontinuous drop in the generalization error, for both binary and continuous weights. The poorly generalizing symmetric states are metastable for arbitrarily large training sets.     

自然循环流动不稳定性的多目标优化极限学习机预测方法

极限学习机是近年来提出的一种前向单隐层神经网络训练算法,具有训练速度快、不会陷入局部最优等优点,但其性能会受到随机选取的输入权值和阈值的影响.针对这一问题,提出一种基于多目标优化的改进极限学习机,将训练误差和输出层权值的均方最小化同时作为优化目标,采用带精英策略的快速非支配排序遗传算法对极限学习机的输入层到隐层的权值和阈值进行优化.将该算法应用于摇摆工况下自然循环系统不规则复合型流量脉动的多步滚动预测,分析了训练误差和输出层权值对不同步长预测效果的影响.仿真结果表明,优化极限学习机预测误差可以用较小的网络规模获得很好的泛化能力.为流动不稳定性的实时预测提供了一种准确度较高的途径,其预测结果可以作为核动力系统操作员的参考.     

基于光谱信息的空间目标模式识别算法研究

在观测空间目标时,往往会受到地基观测仪器等因素的制约,导致无法利用目标图像信息从外形上进行识别。根据不同空间目标表面组成材料不同,其产生的反射光谱会存在差异这一特性,可利用空间目标特有的光谱信息进行识别分类。基于此,从光谱学角度对空间目标识别算法进行研究,在K最近邻算法(KNN)的基础上,采用了一种自适应权重局部超平面方法(AWKH),算法主要在计算预测样本与超平面距离时加入对特征权重的考虑,构建了以样本特征组间差与组内差的比值作为特征权重值的超平面模型,从而提高了分类效果和分类效率。为验证算法的分类效果,本文进行了四组验证实验,第一组实验将美国地质勘探局数据库中提取出的九种常用材料光谱随机选出三种混合成多类进行识别;第二、三组实验将四种常用空间目标材料的光谱作为纯物质光谱,分别从可见光和近红外波段对其混合物质进行分类;第四组实验通过实测四个方形模型样本六个面的光谱对其进行识别分类。实验过程中将实验结果与目前常用的支持向量机(SVM)进行对比,对比结果表明改进后的AWKH算法在识别精度和样本适用范围上具有更高的优越性。     

Optimized convergence for multiple histogram analysis

We propose a new algorithm for solving the weighted histogram analysis method (WHAM) equations to estimate free energies out of a set of Monte Carlo (MC) or molecular dynamics (MD) simulations. The algorithm, based on free-energy differences, provides a more natural way of approaching the problem and improves convergence compared to the widely used direct iteration method. We also study how parameters (temperature, pressure, etc.) of the independent simulations should be chosen to optimize the accuracy of the set of free energies.     

Thermodynamic properties of solvated peptides from selective integrated tempering sampling with a new weighting factor estimation algorithm

Conformational sampling under rugged energy landscape is always a challenge in computer simulations. The recently developed integrated tempering sampling, together with its selective variant (SITS), emerges to be a powerful tool in exploring the free energy landscape or functional motions of various systems. The estimation of weighting factors constitutes a critical step in these methods and requires accurate calculation of partition function ratio between different thermodynamic states. In this work, we propose a new adaptive update algorithm to compute the weighting factors based on the weighted histogram analysis method (WHAM). The adaptive-WHAM algorithm with SITS is then applied to study the thermodynamic properties of several representative peptide systems solvated in an explicit water box. The performance of the new algorithm is validated in simulations of these solvated peptide systems. We anticipate more applications of this coupled optimisation and production algorithm to other complicated systems such as the biochemical reactions in solution.     

A Simple Inductive Approach to the Problem of Convergence of Cluster Expansions of Polymer Models

We explain a simple inductive method for the analysis of the convergence of cluster expansions (Taylor expansions, Mayer expansions) for the partition functions of polymer models. We give a very simple proof of the Dobrushin–Kotecký–Preiss criterion and formulate a generalization usable for situations where a successive expansion of the partition function has to be used.     

基于压缩感知与扩展小波树的自适应压缩成像

为了在现有的采样条件下,通过新的压缩采样方式获得计算量小且质量更好的图像,提出了基于压缩感知与扩展小波树的自适应压缩成像方法。首先将图像投影到分区控制的DMD上,获得图像在低分辨率下的测量值,并通过压缩感知重构算法重构出低分辨图像,接着利用扩展小波树预测重要小波位置,通过DMD在小波域采样获取图像的细节信息,最后由小波逆变换恢复高分辨率图像。将该方法与最小化全变分算法（TVAL3）和近来提出的基于扩展小波树的自适应成像算法（EWT-ACS）效果进行对比,实验结果表明,以boat图像为例,在压缩感知采样率为0.75,整体采样率为10%的无噪声条件下,该方法相较于TVAL3、EWT-ACS算法信噪比提高了4.63 dB和2.87 dB,在附加噪声条件下成像效果也较好。该方法能极大地降低压缩感知重建算法的运行时间,同时减少采样次数,具有较好的抗噪性。     

基于人眼视觉的对不良照明图像的二值化方法

提出一种新的基于视觉特性的自适应阈值分割方法.利用人眼视觉对比敏感度特性把图像块分成两类,分别借助OTSU算法和模拟人眼识别过程的多尺度模糊隶属方法实现对这两类具有不同灰度特性图像块的自动阈值分割.实验结果表明,使用该方法能够有效的克服不良照明的影响,还原图像的原本特征信息,二值化效果较好.     

A Novel Adaptive Feature Fusion Strategy for Image Retrieval

In the era of big data, it is challenging to efficiently retrieve the required images from the vast amount of data. Therefore, a content-based image retrieval system is an important research direction to address this problem. Furthermore, a multi-feature-based image retrieval system can compensate for the shortage of a single feature to a certain extent, which is essential for improving retrieval system performance. Feature selection and feature fusion strategies are critical in the study of multi-feature fusion image retrieval. This paper proposes a multi-feature fusion image retrieval strategy with adaptive features based on information entropy theory. Firstly, we extract the image features, construct the distance function to calculate the similarity using the information entropy proposed in this paper, and obtain the initial retrieval results. Then, we obtain the precision of single feature retrieval based on the correlation feedback as the retrieval trust and use the retrieval trust to select the effective features automatically. After that, we initialize the weights of selected features using the average weights, construct the probability transfer matrix, and use the PageRank algorithm to update the initialized feature weights to obtain the final weights. Finally, we calculate the comprehensive similarity based on the final weights and output the detection results. This has two advantages: (1) the proposed strategy uses multiple features for image retrieval, which has better performance and more substantial generalization than the retrieval strategy based on a single feature; (2) compared with the fixed-feature retrieval strategy, our method selects the best features for fusion in each query, which takes full advantages of each feature. The experimental results show that our proposed method outperforms other methods. In the datasets of Corel1k, UC Merced Land-Use, and RSSCN7, the top10 retrieval precision is 99.55%, 88.02%, and 88.28%, respectively. In the Holidays dataset, the mean average precision (mAP) was 92.46%.     

Sequential trans-dimensional Monte Carlo for range-dependent geoacoustic inversion

This paper develops a sequential trans-dimensional Monte Carlo algorithm for geoacoustic inversion in a strongly range-dependent environment. The algorithm applies advanced Markov chain Monte Carlo methods in combination with sequential techniques (particle filters) to carry out geoacoustic inversions for consecutive data sets acquired along a track. Changes in model parametrization along the track (e.g., number of sediment layers) are accounted for with trans-dimensional partition modeling, which intrinsically determines the amount of structure supported by the data information content. Challenging issues of rapid environmental change between consecutive data sets and high information content (peaked likelihood) are addressed by bridging distributions implemented using annealed importance sampling. This provides an efficient method to locate high-likelihood regions for new data which are distant and ∕ or disjoint from previous high-likelihood regions. The algorithm is applied to simulated reflection-coefficient data along a track, such as can be collected using a towed array close to the seabed. The simulated environment varies rapidly along the track, with changes in the number of layers, layer thicknesses, and geoacoustic parameters within layers. In addition, the seabed contains a geologic fault, where all layers are offset abruptly, and an erosional channel. Changes in noise level are also considered.     

Metrics for ergodicity and design of ergodic dynamics for multi-agent systems

In this paper we propose a metric that quantifies how far trajectories are from being ergodic with respect to a given probability measure. This metric is based on comparing the fraction of time spent by the trajectories in spherical sets to the measure of the spherical sets. This metric is shown to be equivalent to a metric obtained as a distance between a certain delta-like distribution on the trajectories and the desired probability distribution. Using this metric, we formulate centralized feedback control laws for multi-agent systems so that agents trajectories sample a given probability distribution as uniformly as possible. The feedback controls we derive are essentially model predictive controls in the limit as the receding horizon goes to zero and the agents move with constant speed or constant forcing (in the case of second-order dynamics). We numerically analyze the closed-loop dynamics of the multi-agents systems in various scenarios. The algorithm presented in this paper for the design of ergodic dynamics will be referred to as Spectral Multiscale Coverage (SMC).