GMAT: Glottal closure instants detection based on the Multiresolution Absolute Teager–Kaiser energy operator

Glottal Closure Instants (GCIs) detection is important to many speech applications. However, most existing algorithms cannot achieve computational efficiency and accuracy simultaneously. In this paper, we present the Glottal closure instants detection based on the Multiresolution Absolute TKEO (GMAT) that can detect GCIs with high accuracy and low computational cost. Considering the nonlinearity in speech production, the Teager–Kaiser Energy Operator (TKEO) is utilized to detect GCIs and an instant with a high absolute TKEO value often indicates a GCI. To enhance robustness, three multiscale pooling techniques, which are max pooling, multiscale product, and mean pooling, are applied to fuse absolute TKEOs of several scales. Finally, GCIs are detected based on the fused results. In the performance evaluation, GMAT is compared with three state-of-the-art methods, MSM (Most Singular Manifold-based approach), ZFR (Zero Frequency Resonator-based method), and SEDREAMS (Speech Event Detection using the Residual Excitation And a Mean-based Signal). On clean speech, experiments show that GMAT can attain higher identification rate and accuracy than MSM. Comparing with ZFR and SEDREAMS, GMAT gives almost the same reliability and higher accuracy. In addition, on noisy speech, GMAT demonstrates the highest robustness for most SNR levels. Additional comparison shows that GMAT is less sensitive to the choice of scale in multiscale processing and it has low computational cost. Finally, pathological speech identification, which is a concrete application of GCIs, is included to show the efficacy of GMAT in practice. Through this paper, we investigate the potential of TKEO for GCI detection and the proposed algorithm GMAT can detect GCIs with high accuracy and low computational cost. Due to the superiority of GMAT, it will be a promising choice for GCI detection, particularly in real-time scenarios. Hence, this work may contribute to systems relying on GCIs, where both accuracy and computational cost are crucial.     

一种新型的半导体SMA缺陷识别方法的研究

摘 要: 针对半导体器件在封装工艺中出现表面缺陷,及缺陷形态多样性和不可预测性而带来的模型适应性低等问题,提出了基于双向二维主成分分析和改进的卷积神经网络相结合的缺陷识别方法。首先为克服样本不均匀带来的识别精度低问题,对训练图像进行反射变换等操作构造虚拟样本,然后使用Bi-2DPCA对图像进行降维压缩,提取图像主要特征,再由改进的AlexNet网络进行缺陷识别分类,并提出正态随机采样层,将其加在AlexNet网络的卷积层后进行下采样,同时在全连接层中引入DropConnect来提高网络的泛化性能。实验表明,提出的算法较相关算法有较高的识别率,并在实际的SMA塑封图像数据上得到了验证,同时该算法具有较好的泛化性能。     

基于改进空间金字塔池化卷积神经网络的交通标志识别

针对雾天、光照、遮挡和大倾角等因素导致的交通标志识别准确率低、泛化性差等问题,提出一种基于神经网络的轻量级交通标志识别方法。首先,利用图像归一化、仿射变换和限制对比度自适应直方图均衡化（CLAHE）方法进行图像预处理,以提高图像质量;其次,基于卷积神经网络（CNN）,融合空间金字塔结构和批量归一化（BN）方法构建改进空间金字塔池化卷积神经网络（SPPN-CNN）模型,并利用Softmax分类器实现交通标志分类;最后,选用德国交通标志识别数据集（GTSRB）,对比不同图像预处理方法、模型参数和模型结构的训练效果,并验证和测试所提模型。实验结果表明,SPPN-CNN模型的识别精度达到98.04%,损失小于0.1,在低配GPU条件下识别速率大于3 000 frame/s,验证了模型精度高、泛化性强、实时性好的特点。     

Pooling data from fatality analysis reporting system (FARS) and generalized estimates system (GES) to explore the continuum of injury severity spectrum

Fatality Analysis Reporting System (FARS) and Generalized Estimates System (GES) data are most commonly used datasets to examine motor vehicle occupant injury severity in the United States (US). The FARS dataset focuses exclusively on fatal crashes, but provides detailed information on the continuum of fatality (a spectrum ranging from a death occurring within thirty days of the crash up to instantaneous death). While such data is beneficial for understanding fatal crashes, it inherently excludes crashes without fatalities. Hence, the exogenous factors identified as critical in contributing (or reducing) to fatality in the FARS data might possibly offer different effects on non-fatal crash severity levels when a truly random sample of crashes is considered. The GES data fills this gap by compiling data on a sample of roadway crashes involving all possible severity consequences providing a more representative sample of traffic crashes in the US. FARS data provides a continuous timeline of the fatal occurrences from the time to crash – as opposed to considering all fatalities to be the same. This allows an analysis of the survival time of victims before their death. The GES, on the other hand, does not offer such detailed information except identifying who died in the crash. The challenge in obtaining representative estimates for the crash population is the lack of readily available “appropriate” data that contains information available in both GES and FARS datasets. One way to address this issue is to replace the fatal crashes in the GES data with fatal crashes from FARS data thus augmenting the GES data sample with a very refined categorization of fatal crashes. The sample thus formed, if statistically valid, will provide us with a reasonable representation of the crash population.This paper focuses on developing a framework for pooling of data from FARS and GES data. The validation of the pooled sample against the original GES sample (unpooled sample) is carried out through two methods: (1) univariate sample comparison and (2) econometric model parameter estimate comparison. The validation exercise indicates that parameter estimates obtained using the pooled data model closely resemble the parameter estimates obtained using the unpooled data. After we confirm that the differences in model estimates obtained using the pooled and unpooled data are within an acceptable margin, we also simultaneously examine the whole spectrum of injury severity on an eleven point ordinal severity scale – no injury, minor injury, severe injury, incapacitating injury, and 7 refined categories of fatalities ranging from fatality after 30 days to instant death – using a nationally representative pooled dataset. The model estimates are augmented by conducting elasticity analysis to illustrate the applicability of the proposed framework.     

基于云计算的一体化高清视频会议组网研究

结合实际,介绍了国网河南省电力公司行政、应急、卫星通信一体化高清视频会议组网,高清视频会议的设备级联拓扑、带宽分配、卫星通信系统高清化接入。重点阐述了原有一体化平台的缺陷和基于云计算的高清视频会议平台应用的必要性,并对云计算平台总体架构、云管理平台、资源池的构成进行了分析。通过对云平台中硬件设备的虚拟化、MCU级联及动态调配和云平台中会议流程标准化等关键技术进行研究和实现,保障了云平台中高清视频设备的安全、有效、稳定运行。     

Pooling‐Based Data Interpolation and Backdating

Abstract. Pooling forecasts obtained from different procedures typically reduces the mean square forecast error and more generally improve the quality of the forecast. In this paper, we evaluate whether pooling‐interpolated or‐backdated time series obtained from different procedures can also improve the quality of the generated data. Both simulation results and empirical analyses with macroeconomic time series indicate that pooling plays a positive and important role in this context also.     

DNA library screening, pooling design and unitary spaces

Pooling design is an important research topic in bio-informatics due to its wide applications in molecular biology, especially DNA library screening. In this paper, with unitary spaces over finite fields, we present two new constructions whose efficiency ratio, i.e., the ratio between the number of tests and the number of items, is smaller than some of the existing constructions.     

基于深度学习的卫星图像道路分割算法

针对道路分割时存在的梯度消失问题,构建基于U-Net的卫星道路图像语义分割模型。通过密集连接模块减少梯度消失,并引入空间空洞金字塔结构保留更多的图像特征,在学习深层次特征信息时采用注意力监督机制,提取道路要素的特征信息。在卫星图像道路数据集上的测试结果表明,与FCN、SegNet、U_Net算法相比,该算法模型的准确率、召回率和精确率指标分别达到96.3%、96.9%和96.6%,能够有效地对道路元素进行准确分割。     

深度视觉特征与策略互补融合的图像质量评价*

视觉多特征融合方法未考虑图像不同特征之间和不同评价算法之间的视觉互补性.通过融合人类视觉系统前端生理感知和后端心理处理特性,文中提出深度视觉特征互补融合(CPDVF)的图像质量评价方法.CPDVF深度提取图像的多通道直方图统计和多通道梯度结构这2种互补视觉特征,并进行深度视觉处理.然后设计局部失真度评价和局部相似度评价2种互补算法,分别对失真图像的上述互补视觉特征进行评价.最后联合视觉心理特性和回归函数,融合2种特征评价,获得失真图像质量的客观评价.实验表明,相比特征相似度、视觉显著等多特征联合方法,文中方法在准确度、单调性和可靠性指标上优势明显.