关于负超几何分布期望与方差算法的注记

研究了非还原取样模型中负超几何随机变量的联合分布,得到了若干有用的推论.据此给出了负超几何分布的期望和方差的一种分解算法.     

基于视觉辐条图案识别方法的蓝宝石引晶机制研究

工业生产蓝宝石晶体过程中,引晶步骤有着至关重要的地位。引晶必须在温度梯度较小,温度分布趋于稳定的条件下进行。目前,工业生产蓝宝石主要依靠人工经验操控籽晶杆实现引晶操作,但是人工引晶操作的准确性不高会导致成品品质不佳、资源浪费。为此,本文提出一种基于蓝宝石视觉辐条图案识别方法来检测蓝宝石熔体状态自由液面状态,从而实现一种高效率引晶的机制。此方法利用经典骨架化算法细化辐条图案,Harris算子实现特征信息的提取,提取的特征信息放入运动轨迹模型中判断熔体稳定性,分析液面温度分布稳定性从而实现引晶。结果表明,此算法具有有效性,蓝宝石晶体引晶效率大大提高,生产出的成品良率也有提升,可有效指导蓝宝石的工业生产。     

Triterpenes and limonoids of Cedrela: Distribution,biosynthesis, and 1H and 13C NMR data

Cedrela genus, a member of the Meliaceae family, presents both chemical characteristics associated with and those that distinguish it from the rest of its members. The presence of triterpenes and limonoids is the characteristic of the Meliaceae family, but the class and type of these chemical constituents are distinctive for each genus. Cedrela includes cycloartane, ursane, oleanane, tirucallane, butyrospermane, and apotirucallane triterpenes, and its limonoids belongs to six class and nine types, known as class Ia-type havanensines, class Ib-type delevoyin, class II-type gedunin, class IIIb-type andirobin, class IIIg-type mexicanolide, class IVa-type evoludone, class Va-type obacunol, class V-type limonin, and class VIII. Each of these structural arrangements includes specific traits, defined by their biosynthetic origin, which can be established by means of structural elucidation techniques, particularly ¹H and ¹³C NMR, which assisted by 2D NMR techniques, allowing to deduce their structures unequivocally. The constant presence of these skeletal arrangements in Cedrela ensures that they are its chemophenetic markers and their recurrence is an important criterion for their identity. This review is a compilation of the occurrence of triterpenes and limonoids in Cedrela genus, detailing their biosynthetic association and collecting and organizing their NMR data, with the purpose of facilitating its location, analysis, and use in the phytochemical study of species from this genus.     

一种通过梯形PPMgLN波导倍频实现带宽可调谐光源的理论研究

本文设计了一种梯形的周期极化掺镁铌酸锂(PPMgLN)波导,并通过在传播方向上引入温度梯度来拓宽其倍频(SHG)过程的泵浦光源可接收带宽。通过有限差分的光束传输法,计算波导的有效折射率,并进行波导尺寸的设计。结果表明,通过改变梯形波导不同位置的温度,使其形成一个温度梯度,可拓宽泵浦光源的波长可接收带宽。本文所设计的PPMgLN波导最大泵浦光源可接收带宽为C波段,即1 530~1 565 nm,该波导可倍频C波段,得到输出波段带宽为765~782.5 nm,温度调谐范围为30~150 ℃。     

金刚石衬底上GaN HEMT沟道温度建模:各向异性且非均匀热导率的影响

为了改善GaN HEMT的自热效应,集成高热导率的金刚石衬底有助于增强器件有源区的热量耗散。然而,化学气相淀积(CVD)生长的多晶金刚石(PCD)具有柱状晶粒结构,导致了各向异性的材料热导率,且其热导率值与生长厚度有关。为此,通过建模金刚石生长过程中晶粒尺寸的演变过程,计算了金刚石沿面内和截面方向的热导率。基于该PCD热导率模型,利用计入材料非线性热导率的GaN器件热阻解析模型,计算得到了GaN HEMT沟道温度的波动范围,并分析了其与器件结构(栅长、栅宽、栅间距、衬底厚度)和功耗的依赖关系。最后,通过与有限元(FEM)仿真结果对比,分区域提取了GaN HEMT器件中PCD衬底的有效热导率,分别为260~310 W/(m·K)和1 250~1 450 W/(m·K)。本文的计算为预测金刚石衬底上GaN HEMT器件的沟道温度提供了快速、有效的方法。     

Effect of pressure on the low-temperature reaction of ethylene with N2O4

Calorimetric monitoring of the autoclave reaction N₂O₄ + C₂H₄ at –85 to +10 °C under argon pressure 10–30 bar revealed that the exothermic chemical reaction started at temperatures above –52 °C at 10 bar, whereas an intensive exothermic reaction started at –85 °C and pressure of 30 bar. IR study showed that oligo/polynitroethylene was formed at 30 bar, while carbonyl and hydroxy compound as well as nitrate R–ONO₂ formation occurred upon processing at 10 bar.     

The value function of a transportation problem

We investigate the value of an optimal transportation problem with the maximization objective as a function of costs and vectors of production and consumption. The value is concave in production. For generic costs, the numbers of linearity domains and peak points are independent of costs and consumption. The peak points are determined by an auxiliary assignment problem. The volumes of the linearity domains are independent of costs while their dependence on consumption can be expressed via the multinomial distribution.     

热解吸反压低温等离子体电离源质谱检测食品中的邻苯二甲酸酯EI北大核心CSCD

基于低温等离子体设计并实现了一套热解吸反压低温等离子体电离(TD-iLTPI)装置。TD-iLTPI装置由热解吸模块和电离源模块集成在一个π型四通管上组成,进样探针取得样品后插入热解吸模块使其气化,之后随载气进入电离区域被电离。iLTPI内部的针电极接交流电,外部的环形电极接地,电极连接方式与LTP相反。该文对热解吸反压低温等离子体电离源进行了参数的优化,并与三重四极杆质谱联用检测了12种代表性的邻苯二甲酸酯,同时考察了TD-iLTPI-MS的分析性能,并对含有邻苯二甲酸酯的实际样品进行了检测。结果表明,邻苯二甲酸丁苄酯(BBP)的标准曲线具有良好的线性相关系数(R;=0.9958),加标回收率为89.7%~116.8%,相对标准偏差为4.5%~8.2%,对邻苯二甲酸酯的检出限也远低于其他敞开式离子源。对白酒、果汁、面包、奶酪和黄油中的邻苯二甲酸酯进行了快速筛查,并定量检测了果汁中的邻苯二甲酸丁苄酯。     

基于快速神经网络架构搜索的鲁棒图像水印网络算法

为解决深度学习在图像水印算法中计算量大且模型冗余的问题，提高图像水印算法在抵抗噪声、旋转和剪裁等攻击时的鲁棒性，提出基于快速神经网络架构搜索（neural architecture search，NAS）的鲁棒图像水印网络算法。通过多项式分布学习快速神经网络架构搜索算法，在预设的搜索空间中搜索最优网络结构，进行图像水印的高效嵌入与鲁棒提取。首先，将子网络中线性连接的全卷积层设置为独立的神经单元结构，并参数化表示结构单元内节点的连接，预先设定结构单元内每个神经元操作的搜索空间；其次，在完成一个批次的数据集训练后，依据神经元操作中的被采样次数和平均损失函数值动态更新概率；最后，重新训练搜索完成的网络。水印网络模型的参数量较原始网络模型缩减了92%以上，大大缩短了模型训练时间。由于搜索得到的网络结构更为紧凑，本文算法具有较高的时间性能和较好的实验效果，在隐藏图像时，对空域信息的依赖比原始网络更少。对改进前后的2个网络进行了大量鲁棒性实验，对比发现，本文算法在CIFAR-10数据集上对抵抗椒盐噪声和旋转、移除像素行（列）等攻击优势显著；在ImageNet数据集上对抵抗椒盐高斯噪声、旋转、中值滤波、高斯滤波、JPEG压缩、裁剪等攻击优势显著，特别是对随机移除行（列）和椒盐噪声有较强的鲁棒性。