自动发音错误检测中基于F1值最大化的声学模型训练方法

摘要为了提高计算机辅助语言学习中自动发音错误检测系统的性能,提出一种声学模型的区分性训练方法。该方法将经过正确度标注的非母语语音数据库上的发音错误检测的F₁值的最大化作为模型参数的训练准则。采用Sigmoid 函数对F₁值函数进行平滑构造目标函数,并利用构造弱意义辅助函数的方法以及扩展Baum-Welch 形式的参数更新公式进行优化。提出在模型参数更新与音素门限同时优化的策略保证目标函数增长的单调性。发音错误检测实验表明该方法能够有效地增大训练和测试数据检错的F₁值。同时训练数据和测试数据上的精确度、召回率以及检测正确度都有明显改进。      

奥氏体不锈钢焊缝中的声学仿真模型与全聚焦成像检测

声波在奥氏体不锈钢焊缝中传播时声束弯曲,为超声成像带来了困难。基于Ogilvy焊缝模型,建立了奥氏体不锈钢焊缝非均匀各向异性声场仿真模型,采用Dijkstra路径搜索算法对各向异性条件下的声传播路径和声传播时间进行了数值模拟和分析。在此基础上,采用Verasonics超声相控阵成像系统,进行了奥氏体不锈钢焊缝的全聚焦成像实验,采集全矩阵回波数据,并结合理论模型计算的声传播路径和相应的传播时间,进行了成像结果修正。结果表明,与均匀介质模型的全聚焦成像结果相比,基于该文模型的焊缝全聚焦成像检测结果具有更高的缺陷定位精度和分辨率,验证了该方法的可行性,为奥氏体不锈钢焊缝成像检测提供了新的思路。     

柱面镜组的计算机辅助调校

计算机辅助装调技术是通过计算机对理论和实时检测结果的分析计算来有效地指导装调技术。本文通过计算机对由柱面反射镜组成的红外反射扩束系统的各种调整自由度的详细分析 ,以及在检测时对结果的实时分析 ,可确定合理的检测方案 ,有效的减少了调整时间 ,得到了理想的结果     

过程控制系统中多条件约束的多传感器故障检测与诊断

针对工业过程控制系统中的故障具有类型多样、时空独立和非线性等特点,使得检测与诊断效率降低,系统性能下降等问题,提出了一种基于自定义多条件约束的多传感器故障检测与诊断机制。该机制,首先考虑了系统的稳态和时空特征建立了非线性过程控制系统多故障模型,并给出了满足条件判定法则;然后对于系统中的单故障,并发故障和通信故障等类型给出了多条件约束法则及独立特性判断;最后提出了通过自定义多条件约束的多传感器故障检测与诊断机制。实验结果表明,在平均检测概率、稳态特征保持能力和系统功耗等方面明显优于无条件约束的机制,可以显著改善过程控制系统性能。     

基于SPCNN与改进型矢量CV模型的乳腺X射线肿块分割方法

作为乳腺癌计算机辅助诊断系统的重要环节,肿块分割的结果严重影响到肿块良恶性的判别.针对现有方法的不足,本文提出了一种基于简化型脉冲耦合神经网络和改进型矢量无边缘活动轮廓模型的乳腺X射线肿块分割方法.首先,通过数学分析计算SPCNN的相关参数与终止条件,进而利用SPCNN模型分割出肿块的初始轮廓.然后,针对传统CV模型的不足,进行相应的修正得到改进型矢量CV模型.最后,结合SPCNN分割出的初始轮廓,利用改进型的矢量CV模型处理ROI分割出肿块.采用北京大学人民医院乳腺中心提供的临床图像以及DDSM数据库的图像进行对比实验,实验结果表明,本文方法相比较现有方法分割结果更为准确,尤其是在处理东方女性致密性案例时,本文方法更有优势.     

基于主动学习的玉米种子纯度检测模型更新

种子纯度反映种子品种在特征特性方面典型一致的程度,提高种子纯度检测的准确性和可靠性对保证种子的质量具有重要的意义。高光谱图像技术可以同时反映种子的内部特征和外部特征,在农产品无损检测中已经得到广泛应用。利用近红外高光谱图像实现农产品无损检测的实质就是建立光谱信息与农产品品质参数之间的数学模型关系。但光谱信息易受环境、时间的影响,当待测样本的产地或者年份发生改变时光谱信息也随之改变,导致建立的模型的稳定性变差、泛化能力减弱。针对这一问题,采用主动学习算法选择具有代表性的待测样本,最终以添加最少最优的样本数来扩大原模型的样本空间,从而实现模型的快速更新,提高模型的稳定性,同时与基于随机选择算法（RS）和Kennard-Stone算法(KS)的模型更新效果进行比较。实验结果表明：在不同样本集划分比例下（1∶1, 3∶1, 4∶1）,利用主动学习添加40个新样本更新后的2010年的玉米种子纯度检测模型对2011年新样本的预测精度由47%,33.75%,49%提高到98.89%,98.33%,98.33%;利用主动学习添加56个新样本更新后的2011年的玉米种子纯度检测模型对2010年新样本的预测精度由50.83%,54.58%,53.75%提高到94.57%,94.02%,94.57%;同时基于主动学习算法的模型更新效果明显优于RS和KS。因此基于主动学习算法实现玉米种子纯度检测模型的更新是可行的。     

面向近红外光谱定量分析的深度学习建模与模型迁移

近红外光谱分析技术依赖于表征光谱向量和预测目标之间关系的化学计量学方法。然而,样品的光谱由信号和各种噪声组成,传统化学计量学方法较难直接提取光谱的有效特征,并为复杂的预测任务建立具有较强泛用性的校正模型。进一步地,受限于仪器间的差异,在一台仪器上建立的模型应用于另一台仪器时,难以取得相同的定量分析结果。为此,提出了一种基于卷积神经网络和迁移学习的定量分析建模及模型传递方案,以提高模型在单仪器和跨仪器上的预测性能。在卷积神经网络的基础上,一种结合多尺度特征融合和残差结构,名为MSRCNN的先进模型被设计,并在主仪器上展现了卓越的预测能力。然后,设计了四种的基于fine-tune模型迁移策略,将在主仪器上建立的MSRCNN模型迁移到从仪器。在药品和小麦的公开数据集上的实验结果表明,MSRCNN在主仪器上的RMSE和R2分别为2.587,0.981和0.309,0.977,优于PLS,SVM和CNN。在利用30个从仪器的样本微调主仪器建立的模型后,迁移MSRCNN中的卷积层和全连接层的方案取得了最好效果,其RMSE和R2可分别达到2.289,0.982和0.379,0.965。增加参与模型微调的从仪器样本,可进一步提高性能。     

离轴非球面反射镜补偿检验的计算机辅助装调技术研究

利用零补偿器实施离轴非球面元件面形的干涉检测中,为了实现反射镜的高准确度检测,对其干涉结果中的误差信息进行了分析.根据零补偿器的补偿原理,提出一种新的调整误差分离方法,建立了离轴非球面补偿检验的调整误差分离模型,并利用该模型对一块离轴非球面反射镜进行了仿真实验.调整前由调整误差引入的波像差为0.2332λRMS(λ=632.8nm),根据仿真结果调整后的波像差为0.0026λRMS,表明该方法具有较高的准确度,可有效提高检测效率.     

基于计算机辅助的医学图像测量与控制方法研究

医学图像常规检查中,需要对目标图像进行测量以获取相关数据。测量主要集中在对距离,面积,角度以及周长等数据的获取。由于医学检测的严谨性,数据准确,方法简便是必不可少的要求。随着计算机技术的发展,其在医学中的应用也得到了广泛的推崇。计算机辅助的医学图像测量不但替代了传统的手工测量方式,也极大地提高了测量的精度与效率。分析医学图像测量算法,针对如何确定目标区域的问题设计了颜色填充的方法,可有效区分出目标区域图像。为了提高该算法的效率,进一步提出了其改进算法,降低了算法执行次数,提高了系统性能。     

一种基于目标与背景特征分离模型的高光谱目标检测修正算法

高光谱图像立方体数据可以提供成像场景中地物在可见光和近红外波长范围内的空间信息和地物属性诊断的光谱特征信息,在目标检测与识别方面拥有得天独厚的天然优势。然而,基于高光谱图像数据的目标检测也存在一定缺陷,如经典的高光谱目标检测算法仅利用光谱维度信息检测目标,检测模型要么对背景高维特征矩阵构建的准确度不足,要么对背景先验光谱特征的完备性要求较高,导致算法对不同复杂度的检测场景适应性不强。因此,基于计算复杂度较低、参数需求量较少且检测性能较为优异的经典多目标检测算法—多目标约束能量最小化(MCEM),提出了一种基于目标与背景环境特征分离模型的高光谱目标检测修正算法(R-MCEM)。首先,设计了一个与目标形状、尺寸相近的逐像元移动运算窗口,依次计算窗口中的每个像元与窗口内其他像元的光谱距离之和D1,像元与各类目标的光谱距离之和D2。其次,采用获得D1/D2最小值的像元替换窗口内的所有像元值。然后,自左向右、自上而下逐像元移动窗口,重复窗口内每一个像元与目标、背景像元的光谱距离运算,并确定窗口内与背景相似度最高、与目标相似度最低的像元。直到移动运算窗口遍历整个高光谱图像,大幅提升了基于目标与背景...     

Overlapping community detection using a generative model for networks

Detecting overlapping communities is a challenging task in analyzing networks, where nodes may belong to more than one community. Many present methods optimize quality functions to extract the communities from a network. In this paper, we present a probabilistic method for detecting overlapping communities using a generative model. The model describes the probability of generating a network with the model parameters, which reflect the communities in the network. The community memberships of each node are determined based on a probabilistic approach using those model parameters, whose values can be obtained by fitting the model to the network. This method has the advantage that the node participation degrees in each community are also computed. The proposed method is compared with some other community detection methods on both synthetic networks and real-world networks. The experiments show that this method is efficient at detecting overlapping communities and can provide better performance on the networks where a majority of nodes belong to more than one community.     

红外热波技术在装甲装备故障诊断和缺陷检测中的应用

针对装甲装备典型故障诊断操作复杂及缺陷检测效率低下等问题,将国外装备维修中广泛应用的红外热波技术引入到装甲装备维修过程中。简要介绍了红外热波技术的基本原理及特点,系统分析了信噪比和热激励2个因素对检测效果的影响,并成功实现了装备发动机动力不足的故障诊断和装甲板裂纹的缺陷检测。试验结果表明:该技术能够在不解体的情况下实现发动机动力不足故障的准确定位,同时能够在3.5 s之内实现装甲车底板裂纹等缺陷的快速检测。     

火控雷达动态目标探测跟踪建模仿真

火控雷达目标探测跟踪是现代与未来防空作战的首要任务,也是火控系统的关键环节。针对火控雷达连续、自动、快速跟踪目标的特点,首先分析了火控雷达系统闭环连续探测跟踪工作流程,然后利用信号级建模方法,在信号层面上进行了雷达回波信号生成、信号处理与跟踪控制的一体化建模,并给出了各模块模型,最后完成动态火控雷达闭环目标距离、方位角与高低角三维探测跟踪测量过程仿真与分析。仿真结果不仅验证了建模方法与模型逼真有效,也对以后火控雷达跟踪性能评估、数据处理等仿真研究有着一定的参考价值。     

声波远探测测井与地震缝洞精细评价技术研究*

地震勘探探测范围广、空间尺度大,数据采集、处理、解释体系完善,资料运用率高,但纵向分辨率低。声波远探测可实现井周数十米范围内的异常体探测,克服了常规测井方法探测深度浅的局限,具有很高的纵向、周向、径向分辨率,但偶极横波远探测具有180°方位不确定性,径向探测范围不及地震勘探。本研究将两者优势结合,通过对缝洞型地质异常体远探测和地震响应特征进行综合分析,约束远探测方位解释结果,精细刻画地质异常体形态及发育情况,完善远探测测井的横向延伸信息,有效提高远探测解释的精准度,最终形成声波远探测测井与地震一体化缝洞综合评价技术,提升缝洞油气藏钻遇率和开发效率。在塔河等缝洞发育地区进行应用,实现多尺度、全方位和高精度的地质异常体描述与解释,为解决油气勘探过程中由于地震分辨率不足造成的脱靶等现象提供了新的思路。     

An age-dependent model of cavity radiation and its detection

Cavity radiation is modelled as a population point process with age-dependent birth rate. A method of phases is introduced to analyze special types of age dependencies. The conditional life-time of the cavity photon is assumed to be in hypothetical phases, the life-span of the phases being distributed independently and exponentially. The analysis leads to an explicit differential equation for the generating function of the population size. The detection process is analyzed and an explicit expression for the correlation of the counts provided. By an appropriate choice of the parameters, the spectrum corresponding to Gaussian Lorentzian light is recovered. The author felicitates Prof. D S Kothari on his eightieth birthday and dedicates this paper to him on this occasion.     

Small object detection in forward-looking infrared images with sea clutter using context-driven Bayesian saliency model

There are two common challenges for small object detection in forward-looking infrared (FLIR) images with sea clutter, namely, detection ambiguity and scale variance. This paper presents a context-driven Bayesian saliency model to deal with these two issues. By inspecting the camera geometry of the FLIR imaging under the background of sea and sky, we observed that there exists dependency relationship between the locations and scales at which objects may occur, and the context which is defined to be the location of horizon line. Based on this observation, we propose to incorporate contextual information into the basic bottom-up saliency computation, and a unified Bayesian model is developed to achieve this goal. The proposed model is generic and can be potentially applied to other circumstances where context is available for facilitating object detection. Experimental results have demonstrated the effectiveness of our method.     

水声被动目标识别技术挑战与展望

低频水声探测和船舶减振降噪技术发展,使得传统水声目标识别技术性能逐渐下降。该文分析了声呐工作带宽、探测频率、船舶减振降噪给识别技术带来的挑战。针对低频声呐广泛使用的低频线谱识别,研究了低频线谱的识别能力问题;针对智能识别技术发展,研究了深度学习技术在船舶辐射噪声识别中的应用问题,并给出了数据试验结果,文章最后指出了水声被动目标识别技术亟需开展的研究内容和方向。     

New multivariate noise model and data detection using the expectation maximization algorithm

A signal sequence detector in a high areal density recording channel is required to provide robust compensation against unexpected error events. Primarily, a number of error events are caused by media noise and nonlinear distortion. The same problem of signal sequence detection remains to be solved in a future magnetic recording system that comes in predisposed to trend for recording by large-sector size instead of existing single-sector one that consists of 512 information 8-bits bytes. For the above problem, this paper shows the signal estimation method based on statistical inference for such a finite mixture model with known number of degraded noise components. Our signal detection scheme with multivariate autoregressive models for total noise and the expectation maximization algorithm is applied to maximum a posteriori estimation for multivariate mixtures of noise. Furthermore, a non-binary low-density parity-check (LDPC) code is used for an error-correcting code that satisfies the specific run-length limited condition in the proposed system. It shows that the proposed error-correcting and signal detection methods are effective in estimating signal sequences degraded by media noise and in improving the error rate performances with respect to the conventional system using the binary LDPC code and univariate autoregressive model.     

Robust object detection based on deformable part model and improved scale invariant feature transform

We propose an approach to improve the detection results of a generic offline trained detector on frames from a specific video. For two consecutive frames of a video with the object, deformable part model (DPM) detection is performed to get the original detections. Then the image patches corresponding to the detected root box and part boxes were respectively obtained. Thirdly, improved scale invariant feature transform features (SIFT) from those image patches were extracted and matched with the SIFT features by KD-Tree. K-means clustering the angle and scale of matched keypoints to filter out the uncorrected matches and further remove false matches by RANSAC algorithm. Finally, the SIFT_DPM detection result from the matches between image patches of continuous frames was obtained. We focus on methods with high precision detection results since it is necessitated in real application. Extensive experiments with state-of-the-art detector demonstrate the efficacy of our approach.     

Effective and robust infrared small target detection with the fusion of polydirectional first order derivative images under facet model

The robust detection of IR small target acts as one of the key techniques in the infrared search and tracking system (IRSTS). This paper presents a new method of small-target detection which formulates the problem as the detection of Gaussian-like spot. Initially, the amendatory first-order directional derivative (AFODD) based on facet model is applied to get the polydirectional derivative IR images, and the direction information of targets is reserved in these images. Then, the AFODD images are fused together to ensure the robustness and effectiveness of target detection. At last, the Principal Component Analysis (PCA) method is carried out to make targets in the fusion image more prominent, so that they can be extracted out by a simple threshold segmentation. Experiment results show that the presented method performs well even in the IR images with complex backgrounds.