Bivariate models from univariate life distributions: A characterization cum modeling approach

Bivariate life distribution models are of importance for studying interdependent components. We present a generic approach by introducing a new concept of characterized model in stead of a characterized distribution. It strikes a balance between characterization and modeling approaches to eliminate their individual limitations and incorporate their respective strengths. The proposed model, being a characterized one, admits many important properties irrespective of the choice of marginal distributions. The retention of univariate IFR, DFR, IFRA, DFRA, NBU, and NWU class properties in the bivariate setup has been ensured along with some results on series combinations and convolution. No other models, available in the literature, can ensure simultaneous retention of these fundamental and extremely important class properties. © 2004 Wiley Periodicals, Inc. Naval Research Logistics, 2004     

Power series expansions for the distribution and mean value function of a geometric process with Weibull interarrival times

The geometric process is considered when the distribution of the first interarrival time is assumed to be Weibull. Its one‐dimensional probability distribution is derived as a power series expansion of the convolution of the Weibull distributions. Further, the mean value function is expanded into a power series using an integral equation. © 2014 Wiley Periodicals, Inc. Naval Research Logistics, 61: 599–603, 2014     

基于计算机视觉的目标方位测量方法

以显著性标志物中的关键点为目标,探讨了一种基于计算机视觉的目标方位测量方法。在光学成像原理基础上,推算出目标二维成像点坐标与其空间三维坐标之间的映射关系,基于计算机视觉理论建立了测量目标方位的数学模型。考虑透镜畸变,基于Zhang平面标定法完成对摄像机参数的标定。对目标二维图像采用Harris算法进行特征角点检测,并通过亚像素技术对检测的目标特征点进行定位。实验结果表明,该方法能够有效提取被测目标特征角点并获取其像素坐标,用于测量目标相对视觉传感器的方位信息可达到较高精度,具备良好的实用价值。     

稀疏卷积计算高效数据加载与输出缓存策略

针对现有神经网络加速器在处理稀疏神经网络时存在的数据加载效率低、乘加资源利用率低、输出缓存寻址逻辑复杂等问题,提出了稀疏卷积计算高效数据加载与输出缓存策略。将属于同一输入通道的非零输入特征图像数据和非零权重进行全对全乘累加运算,降低了非零数据配对难度,提高了乘加资源利用率;通过采用输入驻留计算,以及密集型循环加载特征图像数据,大幅减少了数据片外调取次数;优化了输出缓存设计,解决了现有方案中存在的输出缓存地址访问争用、存储拥塞等问题。实验表明,与采用类似架构的细粒度脉动加速器相比,在处理单元面积上减少了21.45%;在数据加载速度方面平均提高了117.71%在平均乘法器利用率方面提高了11.25%,达到89%。     

求解卷积公式的一种新方法

基于数形结合方法以及变量代换方法发现计算卷积公式存在某些不足之处,研究卷积公式的计算方法,给出一种新的确定积分限的方法.该方法直观明了、容易理解和掌握并且可操作性强,有效避免作图和分析图形的繁琐,并举例说明了该应用方法.     

一阶交流电路暂态分析

以RC串联电路为例,介绍一阶线性电路交流暂态过程的基本分析方法.通过三种不同方法的比较,归纳出对于不同电路应当采取不同处理办法.     

某些环中卷积的快速计算

本文研究了环中卷积的快速计算问题,讨论了计算域中卷积通常使用的Wino-grad 短卷积算法、快速富里叶变换算法以及多项式变换算法对一般环中卷积计算的可适用性。特别地,对应用广泛的矩阵多项式乘积、矩阵卷积及多项式卷积计算提出了比直接计算快得多的算法。     

曲面拟合法和梯度法相结合的图像亚像素配准算法

针对传统亚像素配准算法存在精度不高、计算复杂的问题,提出了一种曲面拟合法和梯度法相结合的图像亚像素配准算法。采用9点相关系数曲面拟合法对图像进行粗配准,求得一个相对粗略的亚像素配准位置;在两幅图像中选取相同尺寸的子区图像,在粗略的亚像素配准位置基础上,采用梯度法最终获得精确的亚像素配准位置。不同平移关系下的样本图像亚像素配准对比实验结果表明,该算法实现了曲面拟合法和梯度法的优势互补,有效提高了图像配准的精度,最大配准绝对误差由0.17像素降低为0.02像素。     

图像超分辨率卷积神经网络加速算法

为了实现模型的实时和嵌入式运行，提出了一种轻量级的卷积神经网络结构。通过采用较小的滤波器尺寸和引入深度可分离卷积，可大量减少模型参数，提高模型非线性表达能力；在网络末端引入子像素卷积层，直接从原始低分辨率图像学习到高分辨率图像的映射，计算成本为原来的1/k²(k为放大因子)。在Set5数据集上的实验表明，所提模型的速度较经典的图像超分辨率重建算法速度提高了25.8倍，能够在通用GPU上实时运行，峰值信噪比平均提高了0.17 dB，并且参数只有它的35%。     

Nonparametric estimation of some quantities of interest from renewal theory

In this paper, confidence intervals are given for two quantities of importance related to renewal processes. For each quantity, two confidence intervals are discussed. One confidence interval is given for general, all‐purpose use. Another confidence interval is given which is easier to compute, but not of general use. The case where data are subject to right censorship is also considered. Some numerical comparisons are made. © 2003 Wiley Periodicals, Inc. Naval Research Logistics 50: 638–649, 2003.