基于多视角距离像的弹道目标三维微动特征提取

针对现有估计方法提取散射点类型单一以及工程实现较为困难的问题,提出了一种基于组网雷达的多视角距离像提取三维进动特征方法。利用雷达网中各雷达多视角观测弹道目标,提取目标的距离像序列信号,求得最强2散射点在雷达视线上的投影长度,提取目标投影长度随时间变化的关系,利用最小二乘估计对关系式中的参数进行估计。运用组网雷达的多视角特性,根据不同视线方向上不同进动角频率的投影分量,提取目标三维微动特征。计算机仿真验证了提取方法的可行性与有用性,估计效果达到预期目的。     

基于FLAKNN的雷达一维距离像目标识别

由于传统KNN算法在应用于高分辨一维距离像进行目标识别时，存在全局使用固定k值和未考虑各特征分量对分类的影响等不足，使得目标识别性能较差.提出一种改进的KNN算法:FLAKNN.通过提取目标高分辨率一维距离像的尺寸、熵、中心距、不规则度、去尺度特征、对称度等稳定特征，使用Fisher判别分析将所有特征分量投影至低维空间，使不同类别间具备最大可分性；结合相邻样本局部的分布情况和k取值的调整，最终使用少数服从多数的投票原则决定测试样本的类别.结果表明，相对传统KNN算法，该算法进一步提升了识别性能.      

利用导引头信息实现目标脱靶方位识别

提出充分利用导引头提供的信息提高引战配合效率.研究了利用导引头天线测量的视线角速度在弹体的投影进行目标脱靶方向角识别的原理.通过对弹目交会段导弹与目标相对运动关系的分析,得出了识别目标脱靶方向角的数学表达式;针对一种导引头结构设计了可以获得所需识别信息的工程实现方案;给出仿真算例分析脱靶方向角识别的精度.仿真结果表明,识别出的脱靶方向角的精度可以满足引战配合的要求.     

一种有效的步态识别方法

提出一种有效的步态识别算法。对于每个步态序列，首先通过背景减差检测运动目标的轮廓，计算轮廓质心与轮廓边缘的距离并进行归一化，然后使用PCA方法获取特征空间并进行投影，接着使用时空相关和投影质心归一化进行相似性度量，最后通过最近邻法和K近邻法进行分类识别。通过大量的实验分析，取得了较高的识别率。     

基于目标视线角速度的反辐射导弹近炸引信开机策略

为避免反辐射导弹近炸引信因开机过早而出现早炸、误炸的情况,提出了一种利用被动雷达导引头测得的目标视线角速度给出近炸引信开机信号的方法. 给出了利用导引头测角信息计算目标视线角速度的方法,建立了末段弹道上目标视线角速度与弹目距离的统计关系;分别以近炸引信不过晚开机和不过早开机为原则,得到了引信开机时导弹到靶平面距离概率分布函数的上下界. 仿真结果表明:在不能直接命中目标天线的情况下,引信开机时导弹到靶平面的距离主要集中在30～120 m,在可以直接命中目标天线的情况下,引信开机时导弹到靶平面的距离小于30 m.      

基于贝叶斯数据融合的多尺度目标识别

针对红外电厂目标识别问题，提出了基于贝叶斯数据融合的多尺度目标识别方法．该方法研究了前视成像末制导过程中图像目标尺度变化引起的视点角度和特征尺度变化规律，建立了分层次的时空特征模型，根据显著性选取目标特征，采用贝叶斯网络把不同尺度下的显著性特征进行融合，得到正确的识别结果．实验表明，该方法能将多尺度目标的不精确、不完整的特征进行融合处理，从而完成了目标的可靠识别．     

基于投影序列质心的不规则形状识别

以交通标记识别为例,针对不规则形状识别提出一种新的方法。选取不规则形状的边缘点做双向投影变换,根据归一化和中心化后的投影序列数据求得加权质量,从而得出投影序列的质心。由于相似质点系的质心距离相近,因此相似投影序列的质心也相近。以棋盘距离和街区距离的线性组合替代欧氏距离来度量投影序列及其质心的相似性,继而得到不规则形状的相似程度。实验结果表明,该算法在保持较高识别精度的同时,运算速率也比较快。     

用于HRRP多类目标识别的D距离分类器

在雷达自动目标识别（radar automatic target recognition,RATR）领域,为了保证基于高分辨距离像（high-resolution range profile,HRRP）的目标识别算法在进行小样本多类目标识别时仍然具有优异的识别性能,需要提出一种同时具备优异泛化性能与低运算复杂度的识别算法。利用比值计算两个向量之间的比值距离,并将比值距离应用于距离分类器中,称之为D距离分类器。然后利用八类地面目标实测数据将D距离分类器与其他一些RATR统计模型进行比较,分别分析其在小样本与多类目标时的识别精度。最终结果验证出D距离分类器在训练样本有限且多类目标识别时仍然具有优异的泛化性能与很低的运算复杂度。     

基于峰谷点特征的红外目标自动识别方法

研究了红外目标图像峰谷点特征提取的方法,提出了一种目标识别相似性度量准则,证明了这种相似性度量在尺度和小角度旋转等变化情况下的不变性,并在此基础上提出了一种基于目标峰谷点特征的自动目标识别算法．真实场景下长波红外目标图像实验证明了该算法的正确性和有效性．     

基于运动模式的点源目标识别算法的研究

目前常用的点源运动目标识别方法并未考虑运动目标自身运动规律,难以识别目标属性,降低了目标识别的准确性。本文提出三点Hermite插值轨迹匹配识别方法,用以识别点源目标运动模式。该方法首先根据已知目标自身运动规律,建立目标运动轨迹全方位投影曲线标准库;然后,采用三次Hermite插值曲率逼近的方法,获得点源目标实际运动轨迹;最后,利用Hansdorff距离匹配实际目标运动轨迹与曲线标准库轨迹,并根据匹配结果识别点源目标。     

基于仿射不变性特征的视点空间划分

为了简化三维物体的识别过程,提高三维物体识别的识别率,该文利用Multi-scale autoconvolution、Trace变换、Zernike矩3种仿射不变性特征,对飞机、汽车、人等三维物体进行视点空间划分,用尽可能少的不等间隔的三维物体的二维投影图像来表达三维物体,并以此为依据进行三维物体识别。在此基础上提出一种针对不同类型物体的仿射不变性特征提取策略,并建立一个实现三维物体任意姿态识别的软件系统平台,应用Princeton形状标准库中的部分模型对该平台进行测试。结果表明,该方法能够取得较好的识别效果,识别率在90%以上。     

基于仿射不变性特征的视点空间划分

为了简化三维物体的识别过程,提高三维物体识别的识别率,该文利用Multi-scale autoconvolution、Trace变换、Zernike矩3种仿射不变性特征,对飞机、汽车、人等三维物体进行视点空间划分,用尽可能少的不等间隔的三维物体的二维投影图像来表达三维物体,并以此为依据进行三维物体识别。在此基础上提出一种针对不同类型物体的仿射不变性特征提取策略,并建立一个实现三维物体任意姿态识别的软件系统平台,应用Princeton形状标准库中的部分模型对该平台进行测试。结果表明,该方法能够取得较好的识别效果,识别率在90%以上。     

Pattern recognition and active vision in chickens

Recognition of objects or environmental landmarks is problematic because appearance can vary widely depending on illumination, viewing distance, angle of view and so on. Storing a separate image or 'template' for every possible view requires vast numbers to be stored and scanned, has a high probability of recognition error and appears not to be the solution adopted by primates. However, some invertebrate template matching systems can achieve recognition by 'active vision' in which the animal's own behaviour is used to achieve a fit between template and object, for example by repeatedly following a set path. Recognition is thus limited to views from the set path but achieved with a minimal number of templates. Here we report the first evidence of similar active vision in a bird, in the form of locomotion and individually distinct head movements that give the eyes a similar series of views on different occasions. The hens' ability to recognize objects is also found to decrease when their normal paths are altered.     

A network that learns to recognize three-dimensional objects

The visual recognition of three-dimensional (3-D) objects on the basis of their shape poses at least two difficult problems. First, there is the problem of variable illumination, which can be addressed by working with relatively stable features such as intensity edges rather than the raw intensity images. Second, there is the problem of the initially unknown pose of the object relative to the viewer. In one approach to this problem, a hypothesis is first made about the viewpoint, then the appearance of a model object from such a viewpoint is computed and compared with the actual image. Such recognition schemes generally employ 3-D models of objects, but the automatic learning of 3-D models is itself a difficult problem. To address this problem in computational vision, we have developed a scheme, based on the theory of approximation of multivariate functions, that learns from a small set of perspective views a function mapping any viewpoint to a standard view. A network equivalent to this scheme will thus 'recognize' the object on which it was trained from any viewpoint.     

A probabilistic paradigm for handling uncertain objects in GIS by randomized graph algebra

Probability theory faces difficulties when it is applied to describing uncertain objects in geographic information system (GIS). This is mainly due to the fact that an object in GIS is normally described by a series of discrete vertexes. Modeling uncertainty objects should be therefore based on error of the composed vertexes. This type of model is normally complex and relatively difficult to implement because of many unknown factors, such as the number of vertexes of a polygon, error nature of each individual vertex and error correlation among the vertexes. In this paper, a probabilistic paradigm for handling uncertain objects in GIS by randomized graph algebra is presented. The theoretical basis for this paradigm is the randomized graph algebra-a probability theory for graph-which is newly proposed in this study. Classical probability theory is based on numerical algebra and is also an extension of numerical algebra by further defining probability density within a numerical domain. In the same token, this study begins with defining graph algebra as the basis for probability theory for graph. First, we adopt the theory of graph algebra and further refine the theory by defining the modulo operation for graph. As a result, a graph can thereafter be treated as a "number" and operated by "addition", "subtraction" and others. Second, we construct a measure space by generating sigma-algebra and defining measurable function upon it. The measure space becomes a probability space when the measurable function is a probability density function. Third, we propose the probabilistic paradigm for describing and inferring the uncertainty of geometric objects in GIS by applying the developed randomized graph algebra.     

带宽自适应的MeanShift目标跟踪算法

针对传统均值漂移(Mean Shift)目标跟踪算法中核函数带宽缺乏良好自适应调整的缺点,提出了自适应调整核函数带宽的Mean Shift目标跟踪算法.该算法首先采用核函数计算目标颜色特征值的概率密度,在视频当前帧目标的最优位置区域由目标颜色特征概率投影生成目标概率密度分布图;然后根据概率密度零阶矩值调整下一帧跟踪窗口...     

文字识别中特征与相似度度量的研究

在大样本测试集下国内现有成熟的OCR识别软件的首位识别准确率为95%~97%之间,在准确率和方法上仍有提升和改进的空间。提出一种基于概率特征和结构特征融合的自适应文字识别算法,模拟人类学习的模式,通过对训练样本的不断学习去构建汉字在测量空间的概率分布矩阵,然后比对原始图像和标准汉字库中汉字的概率分布矩阵的相似度来达到汉字分类的效果。其中相似度度量准则是从矩阵空间的结构和概率2个角度出发去构建的,充分考虑了结构模式识别和统计模式识别的优缺点。实验结果显示算法在训练样本下的首位识别正确率可以达到99.66%,在1 623张非训练样本文字图像下的首位识别正确率可以达到99.13%,在5 515张非训练样本文字图像下的首位识别正确率可以达到98.57%。可以证明提出的相似度度量方法在文字识别中的有效性。     

基于滑动概率神经网络的早期故障诊断

针对机电设备早期故障难以识别的问题，提出了一种动态的概率密度估计方法——滑动概率神经网络，用以跟踪分析测量信号的概率密度变化过程，及时发现早期故障．该网络以固定不变的抽样集作为第一层，动态滑动的测量信号作为样本层，通过求和层得到抽样集的条件概率密度估计，将样本层内测量信号的概率密度动态地投影到统一的抽样集上．将网络分解成以测量值为中心的子网络，来实现网络的递归运算，并且利用高斯函数的快速衰减特性或使用分段线性函数近似高斯函数，从而提高了网络的计算实时性．通过压缩机喘振过程数据的应用实例，表明该方法能够有效识剐故障的早期征兆．     

Gravitational microlensing by low-mass objects in the globular cluster M22.

Gravitational microlensing offers a means of determining directly the masses of objects ranging from planets to stars, provided that the distances and motions of the lenses and sources can be determined. A globular cluster observed against the dense stellar field of the Galactic bulge presents ideal conditions for such observations because the probability of lensing is high and the distances and kinematics of the lenses and sources are well constrained. The abundance of low-mass objects in a globular cluster is of particular interest, because it may be representative of the very early stages of star formation in the Universe, and therefore indicative of the amount of dark baryonic matter in such clusters. Here we report a microlensing event associated with the globular cluster M22. We determine the mass of the lens to be 0.13(+0.03)(-0.02) solar masses. We have also detected six events that are unresolved in time. If these are also microlensing events, they imply that a non-negligible fraction of the cluster mass resides in the form of free-floating planetary-mass objects.     

低碳钢电化学抛光表面形貌的统计学特性

采用试验和统计学方法研究了低碳钢在电化学抛光过程中，表面轮廓高度分布的概率密度函数和轮廓的自相关函数的变化．结果表明：随着电化学抛光的进行，概率密度函数分布呈现出平坦→陡峭→平坦的变化趋势，自相关函数呈现出随机→规律→随机的变化趋势；在特定的抛光阶段，概率密度函数分布变窄，自相关函数变化有明显的周期性；原始表面和完全电抛光表面呈现较强的随机性特征．