基于最小一乘背景预测的红外小目标检测算法

本文提出了一种基于最小一乘背景预测的红外小目标检测算法.首先在建立最小一乘准则背景预测模型的基础上,根据最小一乘估计的性质,应用线性规划的方法解决最小一乘估计中极值的选取问题;然后将原始图像与预测图像相减得到预测残差图像;最后利用基于二维指数熵的图像阚值选取快速算法进行分割.文中给出了实验结果与分析,并与基于最小二乘背景预测的检测算法作了比较.实验结果表明,本文提出的算法具有更高的检测概率,优于基于最小二乘背景预测的检测算法.     

基于隐核最小二乘回归的逆控制

基于对核特征空间和最小二乘回归算法的深入研究,提出一种新的非线性隐核最小二乘回归算法(HK-LSR)并将其应用于非线性系统的逆学习控制.仿真表明,该方法具有良好的泛化性能,所构造的逆学习控制器具有令人满意的控制性能.     

基于稀疏约束的DCD滑动窗RLS水声信道估计算法研究

水声信道具有稀疏性的特点,因此高精度低复杂度的稀疏信道估计算法对水声通信具有重要意义。基于自适应滤波算法的信道估计问题本质上是线性回归模型参数的求解问题,传统的最小二乘（Least Square,LS）、最小均方（Least Mean Square,LMS）及递归最小二乘（Recursive Least Squares,RLS）算法在估计稀疏信道时不仅复杂度较高,而且在求解线性回归模型时,因忽略自变量的多重共线性而使稀疏信道估计精度降低。针对上述问题,首先,在经典RLS算法的代价函数中加入信道系数的范数对其进行约束,从而提高了稀疏信道估计的精度,然后,采用滑动窗的方式对其代价函数进行处理以减少算法的计算量。在此基础上又引入二分坐标下降（Dichotomous Coordinate Descent,DCD）算法搜索单次迭代中使代价函数最小的解,进一步降低了算法的复杂度。仿真结果表明,文中所提的算法相较于经典算法在估计精度和复杂度方面具有一定的优越性。     

基于FSVM时域背景预测的红外弱小目标检测

本文提出了一种基于模糊支持向量机(FSVM)时域背景预测的红外弱小目标检测方法.首先针对前几帧图像中对应同一位置像素点的灰度值序列,利用模糊支持向量机进行函数拟合,并据此预测下一帧图像在该位置处像素点的灰度值:然后将原始图像与预测图像相减得到预测残差图像,利用基于二维Tsallis-Havrda-Charvat熵的阈值选取快速算法进行分割,并根据小目标运动的连续性和轨迹的一致性进一步分离噪声和小目标.文中给出了实验结果及分析,并与现有的检测红外小目标的空域和时域背景预测算法进行了比较.结果表明,本文提出的算法具有更高的检测概率,明显优于已有的基于背景预测的红外小目标检测算法.     

基于灰色小波核偏最小二乘的传感器网络数据预测融合研究

摘　要：针对如何降低传感器网络中采集的非平稳、非线性信号的数据传输量,提出了一种基于灰色Morlet小波核偏最小二乘(GMWKPLS)的预测融合模型。该模型把灰色模型预测的思想融入到核偏最小二乘(KPLS)中,采用构造的Morlet小波核函数进行数据变换,将输入映射到高维非线性的特征空间,在特征空间中,利用线性偏最小二乘方法构造预测融合模型。通过对齿轮箱断齿工况升速过程中的振动信号进行分析,结果表明,该模型使用滑动窗方法不断更新建模数据进行动态预测,预测精度高,可大大降低数据传输量,获得显著的节能收益。通过与灰色RBF核偏最小二乘(GRBFKPLS)和RBF核偏最小二乘(RBFKPLS)预测模型对比,GMWKPLS性能最佳,预测误差范围在±0.4%以内。     

复杂背景下的红外目标自动跟踪算法

提出了一种非线性边缘检测和Mean Shift方法相结合的红外目标检测与跟踪算法.采用双窗口算子的非线性边缘检测算法具有计算量小、速度快、图像质量好等优点.在边缘检测后的二值图像基础上,利用改进的Mean Shift跟踪算法实施目标跟踪.该跟踪算法融合了计算目标区域局部标准差的信息;利用灰度值和局部标准差的概率密度函数来描述目标;同时选择核函数级联方式进行目标密度估计,从而弥补了仅用灰度信息描述目标特征的不足.实验结果表明,该跟踪算法检测出的复杂背景下红外目标边缘清晰,并且能够准确地对目标实施自动跟踪.     

深度稀疏最小二乘支持向量机故障诊断方法研究

提出一种结合多层结构和稀疏最小二乘支持向量机(Sparse Least Squares Support Vector Machine,SLSSVM)的机械故障诊断方法。该方法构建了多层支持向量机(Support Vector Machine,SVM)结构,首先在输入层利用支持向量机对信号进行训练,学习信号的浅层特征,利用"降维公式"生成样本新的表示,并作为隐藏层的输入,隐藏层支持向量机对新样本训练并提取信号的深层特征,逐层学习,最终在输出层输出诊断结果。针对因多层结构带来算法的复杂度以及运行时间增加的问题,采用最小二乘支持向量机(Least Squares Support Vector Machine,LSSVM)技术,并将稀疏化理论与最小二乘支持向量机结合,通过构造特征空间近似最大线性无关向量组对样本进行稀疏表示并依此获得分类判别函数,有效解决了最小二乘支持向量机稀疏性缺乏的问题。最后,通过滚动轴承故障诊断实验验证了该方法的有效性。     

一种基于图像特征提取的浮选回收率预测算法

针对矿物浮选过程中的一类回收率预测问题,提出了一种基于泡沫图像特征提取的预测算法.该算法采用最小二乘支持向量机(LSSVM)建立预测模型,通过施密特正交化对核矩阵进行简约,利用核偏最小二乘方法(KPLS)进行LSSVM参数辨识,以此构造具有稀疏性的LSSVM,有效地减小了算法的计算复杂度.为检验模型泛化及预测能力,为多个泡沫特征信息引入预测模型,采用泡沫图像特征提取方法提取泡沫颜色、速度、尺寸、承载量及破碎率特征.实验结果表明,该预测算法对浮选回收率具有良好预测效果.     

总体最小二乘正则化算法的载荷识别

载荷识别中存在病态矩阵求逆的不稳定性将导致解严重失真。在总体最小二乘(Total Least Squares TLS)算法的思想上进行Tikhonov正则化,构造载荷识别的目标函数。然后利用共轭梯度(Conjunction Gradient CG)法解算该目标函数的最优化问题,提出一种算法易实现、收敛性能好、存储量小,且能全面考虑随机误差影响的CG-TLS正则化算法。经仿真和试验探讨了传递函数矩阵病态产生的原因,借助条件数优选振动响应点,最终检验CG-TLS正则化算法与常用的两种正则化算法在不同噪声水平时载荷识别的效果。结果表明,CG-TLS正则化算法载荷识别效果最优,与真实值吻合好,并具有良好的鲁棒性。因此,应用CG-TLS正则化算法实现载荷识别极具实际意义。     

非线性动态WPLS航煤干点软测量

针对炼油工业过程存在的多变量、非线性和数据动态性问题,提出一种自回归移动平均模型与径向基函数-加权偏最小二乘相结合的非线性动态建模方法。首先建立基于径向基函数-加权偏最小二乘方法的软测量模型,然后利用自回归移动平均模型对数据进行时序分析校正,将动态误差信息加入到模型中去,实现模型的动态装换。将该方法应用到加氢裂化航煤干点的软测量建模中,从而获得比径向基函数-加权偏最小二乘算法更高的预测精度。     

基于滑窗批处理的低检测概率无源定位

针对低检测概率下的无源定位问题,提出一种基于滑窗批处理的多传感器融合跟踪算法.通过将多个低检测概率的无源传感器组网,实现目标信息的空间积累,有效提高目标的检测概率.利用伪线性估计技术将非线性测量转换为伪线性测量,再对多传感器的伪线性测量进行基于滑窗批处理的最小二乘估计,得到目标局部最优解.仿真验证了算法的有效性和分析了参数的影响.结果表明,该算法能大幅提高无源定位系统对目标的检测概率,且能满足系统对跟踪精度及实时性的要求.     

Dud,A Derivative-Free Algorithm for Nonlinear Least Squares

Derivative-free nonlinear least squares algorithms which make efficient use of function evaluations are important for fitting models defined by systems of nonlinear differential equations. A new Gauss-Newton-like algorithm with these properties is developed. The performance of the new algorithm (called Dud for “doesn't use derivatives”) is evaluated on a number of standard test problems from the literature. On these problems Dud competes favorably with even the best derivative-based algorithms.     

Dynamic target tracking with fringe-adjusted joint transform correlation and template matching

Target tracking in forward-looking infrared (FLIR) video sequences is a challenging problem because of various limitations such as low signal-to-noise ratio (SNR), image blurring, partial occlusion, and low texture information, which often leads to missing targets or tracking nontarget objects. To alleviate these problems, we developed a novel algorithm that involves local-deviation-based image preprocessing as well as fringe-adjusted joint-transform-correlation--(FJTC) and template-matching--(TM) based target detection and tracking. The local-deviation-based preprocessing technique is used to suppress smooth texture such as background and to enhance target edge information. However, for complex situations such as the target blending with background, partial occlusion of the target, or proximity of the target to other similar nontarget objects, FJTC may produce a false alarm. For such cases, the TM-based detection technique is used to compensate FJTC breaking points by use of cross-correlation coefficients. Finally, a robust tracking algorithm is developed by use of both FJTC and TM techniques, which is called FJTC-TM technique. The performance of the proposed FJTC-TM algorithm is tested with real-life FLIR image sequences.     

基于KPLS的蒸发器出口溶液浓度软测量方法

针对实际生产过程中结疤、结垢等原因造成的蒸发器出口物料浓度在线检测不准的问题,在分析蒸发器非线性特性及影响出口物料浓度因素的基础上,建立了基于核偏最小二乘法的出口物料浓度的软测量模型,利用KPLS有效的非线性特征提取功能,建立了相关直接检测变量与出口物料浓度之间的非线性关系,实现了出口浓度的在线测量。基于工业数据的仿真结果证明,该方法比线性偏最小二乘法更有效,模型精度满足实际生产工艺要求。     

Elastic net grouping variable selection combined with partial least squares regression (EN-PLSR) for the analysis of strongly multi-collinear spectroscopic data

In this paper a novel wavelength region selection algorithm, called elastic net grouping variable selection combined with partial least squares regression (EN-PLSR), is proposed for multi-component spectral data analysis. The EN-PLSR algorithm can automatically select successive strongly correlated prediction variable groups related to the response variable using two steps. First, a portion of the correlated predictors are selected and divided into subgroups by means of the grouping effect of elastic net estimation. Then, a recursive leave-one-group-out strategy is employed to further shrink the variable groups in terms of the root mean square error of cross-validation (RMSECV) criterion. The performance of the algorithm with real near-infrared (NIR) spectroscopic data sets shows that the EN-PLSR algorithm is competitive with full-spectrum PLS and moving window partial least squares (MWPLS) regression methods and it is suitable for use with strongly correlated spectroscopic data.     

海空复杂背景下红外弱点目标检测新算法

为解决海空复杂背景下红外弱点目标的检测,提出了多量级多向梯度表决融合检测算法。算法依据目标红外辐射特征是像素灰度在水平和垂直方向上梯度变化,将弱点目标特性转化为对图像奇异性的分析。算法用多个量级的梯度步长对图像目标进行检测,并对检测的结果进行表决融合。实验结果表明,红外阈值系数选取2.0-2.4时,算法可对信杂比为1的点目标实现高于95%的检测概率及较低的虚警率。     

Background estimation in nonlinear image restoration

One of the essential ways in which nonlinear image restoration algorithms differ from linear, convolution-type image restoration filters is their capability to restrict the restoration result to nonnegative intensities. The iterative constrained Tikhonov-Miller (ICTM) algorithm, for example, incorporates the nonnegativity constraint by clipping all negative values to zero after each iteration. This constraint will be effective only when the restored intensities have near-zero values. Therefore the background estimation will have an influence on the effectiveness of the nonnegativity constraint of these algorithms. We investigated quantitatively the dependency of the performance of the ICTM, Carrington, and Richardson-Lucy algorithms on the estimation of the background and compared it with the performance of the linear Tikhonov-Miller restoration filter. We found that the performance depends critically on the background estimation: An underestimation of the background will make the nonnegativity constraint ineffective, which results in a performance that does not differ much from the Tikhonov-Miller filter performance. A (small) overestimation, however, degrades the performance dramatically, since it results in a clipping of object intensities. We propose a novel general method to estimate the background based on the dependency of nonlinear restoration algorithms on the background, and we demonstrate its applicability on real confocal images.     

A sub-region image registration algorithm with weight-based hierarchical importance sample consensus for moving target detection

The image registration is the key part in moving target detection (MTD), including extracting feature points, matching feature points and estimating motion model. While MTD always confronts with camera irregular movement and complex scenes where many small moving targets are mixed with the background, the accuracy of motion model estimation is seriously affected by these problems. So, our job focuses on estimating a motion model. We present a novel sub-region image registration algorithm with weight-based hierarchical importance sample consensus (WHISAC) to efficiently estimate motion model. Image registration based on sub-region is for removing background motions in different regions called sub-region, and the motion models of sub-regions are separately estimated. Then, in WHISAC, for overcoming the flaws that the ratios of outliers' increases in sub-regions and small moving targets are regarded as inliers wrongly, weights combining into a quality function are proposed to indicate the types of correspondences for guiding sampling. Samples in the WHISAC are drawn hierarchically a different number in the data set sorted by the quality function. Finally, the motion model based on the consensus is obtained by WHISAC. Compared to RANSAC and LMeds, WHISAC has more accurate consensus, faster convergent speed and different images with less interference which facilitates the subsequent target detections in the MTD system.     

Three‐dimensional support function estimation and application for projection magnetic resonance imaging

Magnetic resonance imaging (MRI) of nuclei that have very short relaxation times is conveniently based on spherical sampling. We have presented a least squares framework for reconstructing three‐dimensional (3D) source distribution images from such data. In this paper, we describe a practical algorithm for 3D support function estimation, which forms the basis for a method called focus of attention. By essentially identifying and eliminating equations and unknowns that merely represent background data, this data‐driven preprocessing scheme effectively reduces the computational burden associated with our algebraic approach to projection MRI. © 2002 John Wiley & Sons, Inc. Int J Imaging Syst Technol 12, 43–50, 2002