基于线性均方误差估值的LPC距离量度

本文从LPC系数的统计特性出发,基于线性均方误差估值的思想,提出了一种用于语音识别的新的LPC距离量度——LMSE距离.在一个IBM-PC机语音处理系统和一个VAX-11/750机语音处理系统上,做了大量孤立词识别实俭,证实了这种距离的可行性.文中还将它与著名的Itakura距离进行了比较.     

Hardware Euclidean Distance Transform Implementation

Based on an efficient algorithm of Euclidean distance transform for binary images, a circuit of O (N^2) size is proposed. With in-place calculation, both the intermediate data storing and the result output use the same menmory with the input data. This reduces the amount of menmory largely. By replacing multipliers with coumters, comparators, and adders, the circuit size is further reduced and its calculation speed is improved also.     

公司债券违约率的结构化模型研究

采用期权定价和违约率函数的结构化处理方法建立估计公司债券违约率的模型，应用美国国内公司的整体数据给出了将模型具体化的处理方法，得到一个带马尔可夫链的违约率函数。根据实证数据的初步检验，该函数的拟合效果比较理想，可用于估计一般性上市公司群体的债券违约率。文中的处理方法具有良好的可行性，只要拥有模型必需的股权价值、债务价值和违约率的数据，就能按照该方法将模型具体化，拟合出违约率函数，并由此来估计公司债券的违约率。     

A Target Recognition Approach to Projecting HRR Profiles onto Subspace

Abstract: A array of the azimuthally averaged range-profile vectors and the inter-class and intra-class divergence matrixesare constructed iwth many frames of the high resolution range profiles which result from radar echoes of airplanes. Takingthe methods of whitening transformation and SVD produces a system of subspace vectors for target recognition. Where-upon, a template library for target recognition is built by the projection of a class-mean vector made from the radar dataonto the subspace for recognition. By Euclidean distance, a comparison is made between the above projection and eachtemplate in the library, to decide which class the target belongs to. Finally, simulations with the experimental radar dataarte given to show that the proposed method is robust to variation in azimuth and immune to additive Gaussian noisewhen SNR≥5dB.     

基于校园网络的远程教育系统方案设计

本文分析了大理学院跨校区的远程教育系统实施方案,设计了一套技术先进、性能可靠、标准开放、扩展灵活的远程教育系统。     

求最佳生物序列并置排列的算法和程序

本文提出了计算最佳生物序列并置排列的程序和算法。程序使用计算进化距离的改进算法,在IBM-PC/AT上用Turbo-Pascal 4.0版开发。程序有友好的用户界面、操作简便、通用性强,可以移植到多种计算和操作环境。本文以蛋白质分子序列为例,给出了复杂赋权情形下计算进化距离和最佳并置排列的算法,证明了不同赋权方式本质上不改变改进算法的时间和空间复杂性。     

一种基于距离比值的支持向量机增量训练算法

由于支持向量机具有较好地学习性能和泛化能力,目前已经得到了广泛的应用。如何使支持向量机进行有效的增量学习是目前支持向量机应用中需要解决的问题。深入研究了支持向量分布特点,提出了一种新的支持向量机增量训练淘汰机制——距离比值算法。该算法根据遗忘规则,设定一个合适的参数,按距离比值法中的定义计算各个样本中心距离与其到最优分类面距离的比值,舍弃对后续训练影响不大的样本,即可对训练数据进行有效的淘汰。对标准数据集的实验结果表明,使用该方法进行增量训练在保证分类精度的同时,能有效地提高训练速度。     

模糊熵、贴近度与距离测度及其内在联系探讨

给出了模糊熵的结构表达式,在介绍模糊贴近度与距离测度概念的基础上,对其性质进行了探讨,揭示了模糊度、贴近度和距离测度三者之间的内在联系。     

High resolution imaging using scanning ion conductance microscopy with improved distance feedback control

Microscopy is an essential technique for observation on living cells. There is currently great interest in applying scanning probe microscopy to image-living biological cells in their natural environment at the nanometer scale. Scanning ion conductance microscopy is a new form of scanning probe microscopy, which enables non-contact high-resolution imaging of living biological cells. Based on a scanned nanopipette in physiological buffer, the distance feedback control uses the ion current to control the distance between the pipette tip and the sample surface. However, this feedback control has difficulties over slopes on convoluted cell surfaces, which limits its resolution. In this study, we present an improved form of feedback control that removes the contribution of up to the third-order slope from the ion current signal, hence providing a more accurate signal for controlling the distance. We show that this allows faster and lower noise topographic high-resolution imaging.     

基于相对编辑相似度的近似重复视频检索和定位

为有效对近似重复视频进行检索和定位,提出了一种基于相对编辑相似度的检索和定位算法.算法包括基于局部特征的视频编码和基于相对编辑相似度的在线检索和定位两部分.基于局部特征的视频编码首先提取数据库视频的关键帧,然后在关键帧中提取Root-SIFT特征描述符并应用层次K-Means聚类算法构建词典,之后将关键帧量化至词袋模型的单词并编码.基于相对编辑相似度的在线检索和定位首先对查询视频进行编码,然后应用相对编辑相似度算法,筛选近似重复视频并对近似重复片段进行定位.实验结果表明,LD算法比Yeh等提出的算法在平均F₁评价准则上效果要高8.55%,并且NDCR降低为原来的29%,效果提升明显.