多尺度卷积神经网络检测睡眠呼吸暂停

睡眠呼吸暂停综合征作为一种常见的与睡眠相关的呼吸障碍性疾病,受到众多的关注。由于其复杂的检诊断过程及昂贵的价格,吸引了众多研究学者探索基于单通道信号的快速、便捷检测方法。基于心电信号(ECG)提出了一种多尺度卷积神经网络睡眠呼吸暂停快速检测方法,与常规的单尺度卷积神经网络方法相比,该方法可以有效地结合信号的细节信息和抽象信息,提升卷积神经网络的特征呈现能力。通过PhysioNet提供的Apnea-ECG数据库进行验证,多尺度卷积神经网络获得了85.2%准确率、83.1%敏感性和86.5%特异性。与现有方法相比,该方法进一步提升了睡眠呼吸暂停的检测性能。     

Paleyfly:一种可扩展的高速互连网络拓扑结构

高速互连网络是高性能计算系统的重要组成部分.随着网络规模需求的扩大,如何搭建更大规模的网络是高速互连网络拓扑结构设计的关键.因此,提出一种新型层次化的拓扑结构Paleyfly (PF),其结合了Paley图强正则的特性和Random Regular(RR)图支持任意规模大小的特点.相比其他新型高速互连网络拓扑结构,Paleyfly能够有效解决在路由芯片端口数受限的背景下,Dragonfly (DF)可扩展性受限、Fat tree(Ft)物理成本高、RR结构物理布局难、路由表规模大等问题.同时,根据强正则属性在路由策略上负载均衡的优势,提出了4种路由策略来解决网络的拥塞问题.最后,通过模拟器实验比较分析PF结构与其他拓扑结构及PF结构不同路由策略的性能,验证了PF结构在不同规模以及不同通信模式配置下网络延迟优于RR结构.     

Integration of independent component analysis and neural networks for ECG beat classification

In this paper, we propose a scheme to integrate independent component analysis (ICA) and neural networks for electrocardiogram (ECG) beat classification. The ICA is used to decompose ECG signals into weighted sum of basic components that are statistically mutual independent. The projections on these components, together with the RR interval, then constitute a feature vector for the following classifier. Two neural networks, including a probabilistic neural network (PNN) and a back-propagation neural network (BPNN), are employed as classifiers. ECG samples attributing to eight different beat types were sampled from the MIT-BIH arrhythmia database for experiments. The results show high classification accuracy of over 98% with either of the two classifiers. Between them, the PNN shows a slightly better performance than BPNN in terms of accuracy and robustness to the number of ICA-bases. The impressive results prove that the integration of independent component analysis and neural networks, especially PNN, is a promising scheme for the computer-aided diagnosis of heart diseases based on ECG.     

室性早搏的多模板匹配自适应识别算法

为实现室性早搏的自动判别,提出一种基于多模板匹配的自适应识别算法。采用滤波提高数据信噪比,利用样本数据通过特定的自学习机制建立模板库,设计模板的压缩算法实现待测心搏与模板心搏宽度的对齐,采用模板队列匹配和模板库匹配相结合的匹配策略,设计自适应更新规则实时优化队列与库的结构,利用相关系数检测室早。使用MIT-BIH数据库进行测试,达到99.39%的灵敏度和98.79%的特异度。     

结合深度学习与条件随机场的遥感图像分类

目的 为进一步提高遥感影像的分类精度,将卷积神经网络（CNN）与条件随机场（CRF）两个模型结合,提出一种新的分类方法。方法 首先采用CNN对遥感图像进行预分类,并将其类成员概率定义为CRF模型的一阶势函数;然后利用高斯核函数的线性组合定义CRF模型的二阶势函数,用全连接的邻域结构代替常见的4邻域或8邻域;接着加入区域约束,使用Mean-shift分割方法得到超像素,通过计算超像素的后验概率均值修正各像素的分类结果,鼓励连通区域结果的一致性;最后采用平均场近似算法实现整个模型的推断。结果 选用3组高分辨率遥感图像进行地物分类实验。本文方法不仅能抑制更多的分类噪声,同时还可以改善过平滑现象,保护各类地物的边缘信息。实验采用类精度、总体分类精度OA、平均分类精度AA,以及Kappa系数4个指标进行定量分析,与支持向量机（SVM）、CNN和全连接CRF相比,最终获得的各项精度均得到显著提升,其中,AA提高3.28个百分点,OA提高3.22个百分点,Kappa提高5.07个百分点。结论 将CNN与CRF两种模型融合,不仅可以获得像元本质化的特征,而且同时还考虑了图像的空间上下文信息,使分类更加准确,后加入的约束条件还能进一步保留地物目标的局部信息。本文方法适用于遥感图像分类领域,是一种精确有效的分类方法。     

The role of oxidation damage in fatigue crack initiation of an advanced Ni-based superalloy

The effects of prior oxidation on the room temperature fatigue life of coarse-grained Ni-based superalloy, RR1000, have been investigated. High cycle fatigue tests were conducted, on both machined and pre-oxidised testpieces, at room temperature at an R ratio of 0.1. The oxidation damage was produced by pre-exposures at 700 °C for either 100 or 2000 h. Pre-oxidised testpieces tended to fail with shorter fatigue lives than those obtained from the as-machined testpieces although they were also observed to outperform the as-machined test pieces at peak stress levels around 900 MPa. The chromia scale and intergranular alumina intrusions formed during pre-oxidation are prone to crack under fatigue loading leading to early crack nucleation and an associated reduction in fatigue life. This has been confirmed to be the case both below and above a peak stress level of ∼900 MPa. The better fatigue performance of the pre-oxidised specimens around this stress level is attributed to plastic yielding of the weaker γ′ denuded zone, which effectively eases the stress concentration introduced by the cracking of the chromia scale and intergranular internal oxides. This γ′ denuded zone is also a product of pre-oxidation and develops as a result of the selective oxidation of Al and Ti. Over a limited stress range, its presence confers a beneficial effect of oxidation on fatigue life.     

全IP移动通信网中的微移动性管理方案

在下一代的全IP移动通信网中,移动性管理是一个非常关键的问题。本文首先阐述了全IP移动通信网中的微移动性管理的概念;其次介绍了在全IP移动通信网中的几种典型的微移动性管理方案,包括CIP方案、HAWAII方案、MIP-RR方案和微移动MPLS方案,并对它们进行了性能比较。     

利用BGP虚拟下一跳技术实现IP地址快速封堵

通过在BGP的路由反射器（RR）把需要封堵的IP地址通过BGP信息向全网发布并携带特定的BGP虚拟下一跳信息,而在IP骨干网的边缘路由器设置该特定虚拟下一跳IP地址的黑洞路由。以至于在IP骨干网的边缘路由器都会自动把访问该需要封堵IP地址的流量丢弃,达到快速封堵IP地址的目的。     

Production of an exopolysaccharide-containing whey protein concentrate by fermentation of whey

Using whey as a fermentation medium presents the opportunity to create value-added products. Conditions were developed to partially hydrolyze whey proteins and then ferment partially hydrolyzed whey with Lactobacillus delbrueckii ssp. bulgaricus RR (RR; an EPS-producing bacterium). In preliminary experiments, pasteurized Cheddar cheese whey was treated with Flavourzyme to partially hydrolyze the protein (2 to 13% hydrolyzed). Fermentation (2 L, 38 degrees C, pH 5.0) with RR resulted in EPS levels ranging from 95 to 110 mg of EPS per liter of hydrolyzed whey. There were no significant differences in the amount of EPS produced during fermentations of whey hydrolyzed to varying degrees. Since a high level of hydrolysis was not necessary for increased EPS production, a low level of hydrolysis (2 to 4%) was selected for future work. In scale up experiments, whey was separated and pasteurized, then treated with Flavourzyme to hydrolyze 2 to 4% of the protein. Following protease inactivation, 60 L of partially hydrolyzed whey was fermented at 38 degrees C and pH 5.0. After fermentation, the broth was pasteurized, and bacterial cells were removed using a Sharples continuous centrifuge. The whey was then ultrafiltered and diafiltered to remove lactose and salts, freeze-dried, and milled to a powder. Unfermented hydrolyzed and unhydrolyzed whey controls were processed in the same manner. The EPS-WPC ingredients contained approximately 72% protein and 6% EPS, but they exhibited low protein solubility (65%, pH 7.0; 58%, pH 3.0).     

Retreatment of silicon slurry by membrane processes

The purpose of the present study is to develop a process to regenerate the polish liquid used in Chemical and Mechanical Polishing (CMP), called “slurry”, and more specifically Silicon CMP slurry. Physico-chemical analyses show a considerable dilution of slurry through washing waters used in polishing. Thus, this effluent has been characterised for a better identification of the deviations from the slurry of reference (Point Of Use). Hence, the principle is to regenerate this effluent by membrane processes. The ultrafiltration results obtained at laboratory scale have led to the development of an industrial prototype. An optimal utilisation of this treatment allows completing a two-step process: the reconcentration by ultrafiltration and a chemical adjustment by addition of concentrated slurry. A stable behaviour of the slurry at the different steps of the process has been observed. Polishing results are similar with retreated and POU slurries. Furthermore, the functioning at industrial scale permits to maintain the performances obtained on the laboratory pilot.