基于双层采样主动学习的社交网络虚假用户检测方法

社交网络的飞速发展给用户带来了便捷,但是社交网络开放性的特点使得其容易受到虚假用户的影响.虚假用户借用社交网络传播虚假信息达到自身的目的,这种行为严重影响着社交网络的安全性和稳定性.目前社交网络虚假用户的检测方法主要通过用户的行为、文本和网络关系等特征对用户进行分类,由于人工标注用户数据需要的代价较大,导致分类器能够使用的标签样本不足.为解决此问题,本文提出一种基于双层采样主动学习的社交网络虚假用户检测方法,该方法使用样本不确定性、代表性和多样性3个指标评估未标记样本的价值,并使用排序和聚类相结合的双层采样算法对未标记样本进行筛选,选出最有价值的样本给专家标注,用于对分类模型的训练.在Twitter、Apontador和Youtube数据集上的实验说明本文所提方法在标签样本数量不足的情况下,只使用少量有标签样本就可以达到与有监督学习接近的检测效果;并且,对比其他主动学习方法,本文方法具有更高的准确率和召回率,需要的标签样本数量更少.     

化学工业绿色制造产业链接技术和发展方向探析

"十三五"以来,我国化学工业的绿色发展取得了显著进步,在产品产量提高的同时,资源消耗、能源消耗、污染排放均实现了大幅减少。绿色制造是当前我国实现可持续发展的必要途径。要实现化学工业的绿色发展,除了要关注化工行业自身的绿色制造,还要关注化工产业与其他行业、与社会进行产业链接时的绿色协调发展。本文在对我国化学工业绿色发展现状进行分析的基础上,给出我国化工行业绿色发展的方向,提出化学工业绿色制造产业链接的关键技术,并通过五个企业案例进行典型模式分析。     

Hydrogen redistribution under stress-induced diffusion and corresponding fracture behaviour of a structural steel

ABSTRACT

Hydrogen redistribution under stress-induced hydrogen diffusion and corresponding fracture behaviour of a 960?MPa grade martensitic steel were studied. Slow strain rate tensile (SSRT) tests after hydrogen pre-charging were performed and the fracture surface was observed and analysed. The strain rate ranged from 10^?6 to 10^?4?s^?1. In the pre-charged sample with a certain hydrogen content of 0.62?ppm, hydrogen distribution was homogeneous before the SSRT test. After tensile testing, brittle fracture features appeared in the centre of the fracture surface, while ductile features appeared in the surrounding area. Brittle region size increased with the strain rate slowing down in the range from 10^?4 to 5?×?10^?6?s^?1, while it stabilised at the strain rate slower than 5?×?10^?6?s^?1. Relationship between the strain rate and the brittle region size was established and discussed based on the present data of hydrogen content in the material.

This paper is part of a thematic issue on Hydrogen in Metallic Alloys     

氢化碳化硅(SixC1-x∶H)薄膜发光特性和纳米线发光增强研究

氢化碳化硅薄膜作为一种宽带隙的半导体材料,具有优越的物理特性,其在光电子器件上的潜在应用引起了人们的兴趣。利用等离子增强化学气相沉积(PECVD)系统制备了一系列氢化碳化硅薄膜,通过改变反应前驱物及流量比调节薄膜的室温光致发光性质。实验发现在一定范围内随着流量比R(CH_4/SiH_4)的提高,氢化碳化硅薄膜的光致发光峰位蓝移且发光强度增强;同时反应前驱物中的氢会极大影响氢化碳化硅薄膜的发光强度。通过椭偏仪(Ellipsometer)测量了薄膜的光学常数,发现薄膜沉积速率随着流量比R的增加而降低;傅里叶红外光谱仪(FTIR)测试表明Si-C有序度随着流量比的增加而增大。同时研究了三维纳米线结构对多态碳化硅薄膜发光性质的影响。光致发光测试结果表明三维纳米线结构可以有效提高薄膜的光致发光强度。     

二维光子晶体微腔的制备及其对硅光学材料的光量子放大

光子晶体(PC)可以增加光物质相互作用和光发射效率,在微纳光子学、量子光学及信息光学等领域中都有着广泛的应用。近年来,二维硅基光子晶体微腔的发光增强效应研究取得了较为重大的突破。本文针对现有二维光子晶体及微腔的制备方法与发光性能的调控展开论述,详细介绍了二维光子晶体微腔的制备进程与温度、泵浦能量、微腔结构对微腔Q因子以及发光性能的影响,并进一步展望了二维光子晶体在硅材料光量子放大领域未来研究所面临的问题及应用前景。     

壳聚糖-二氧化硅复合湿凝胶的改性及其对木家具中有害物质吸附性能的研究

采用改性共溶胶-凝胶法制备多孔壳聚糖气凝胶通过扫描电镜、红外、X射线能谱、BET等手段对其进行分析。结果表明:制备的壳聚糖-二氧化硅复合气凝胶既保留了气凝胶的轻质结构和介孔特性,又含有壳聚糖所具有的烷基、羟基、氨基、酰胺基等活泼有机基团,且通过改性提高其活性,使其能吸附木家具中挥发性有机物和重金属。其中以戊二醛改性的壳聚糖-二氧化硅复合湿凝胶吸附能力最好,其对乙酸丁酯和正十一烷的吸附率为85%~95%,对重金属的吸附率大于50%,其吸附VOCs的能力为二氧化硅凝胶的1.92倍,吸附重金属能力为二氧化硅凝胶的6.86倍。     

Combining differential SAR interferometry and the probability integral method for three-dimensional deformation monitoring of mining areas

With the exploitation of coal resources, ground surface subsidence continues to occur in mining areas, destroying the ecological environment and significantly affecting the daily productivity and life of humans. The differential synthetic aperture radar interferometry (D-InSAR) technique is widely used to monitor ground surface deformation because of its unique advantages such as high accuracy and wide coverage. However, conventional D-InSAR technology provides only one-dimensional (1D) displacement monitoring along the radar line of sight (LOS). This article proposes a method based on an analysis of the mining subsidence law for true three-dimensional (3D) displacement monitoring by combining D-InSAR and a subsidence prediction model based on the probability integral method. In this approach, 1D displacement, obtained using D-InSAR, is then combined with the prediction model to obtain the 3D displacement of ground surface target points. Here, 3D displacement curves were obtained for the Fengfeng mining area (China) using RadarSat-2 images obtained on 9 January and 2 February 2011. True ground surface displacement was measured simultaneously by levelling when the 152under31 s working face was being exploited in Jiulong mine. Vertical displacement and inclined deformation calculated using the proposed method were compared with levelling survey data and the results showed average differences of 3.2 mm and 0.1 mm m^?1, respectively; the calculated maximum displacement in the east–west and south–north directions were 106 and 73 mm, respectively. The spatial distribution of the displacements was in accordance with the mining subsidence law. Thus, the new method can retrieve highly accurate 3D displacements caused by mining subsidence.     

Risk assessment for wildfire occurrence in high-voltage power line corridors by using remote-sensing techniques: a case study in Hubei Province,China

Wildfires are a significant problem for power line maintenance that can lead to widespread power outages and economic loss in China. Remote sensing (RS) and geographic information system (GIS) technologies and their derived applications are effective tools for power line surveillance and disaster prevention. Using multi-source data and remote-sensing techniques, a risk assessment of wildfire occurrence in high-voltage power line corridors in Hubei Province, China, is presented in this study. Both natural and human causes are considered in the assessment, including a traditional Chinese spring custom. Historical ignition points along power line corridors in Hubei Province that occurred from 2009 to 2014 were collected as training data to create a quantitative analysis. Next, a logistic regression model was applied, and ignition probability maps were produced for power lines. Several wildfire accidents that occurred in 2015 were used as validation data to evaluate the performance of the proposed method. The experimental results indicated that the drought conditions significantly influenced wildfires as natural cause, and human activities play important roles in causing wildfires during specific time periods. The effectiveness and robustness of our method is thereby demonstrated, and our method can be used to provide valuable suggestions for wildfire management for the electrical department in Hubei Province.     

深层神经网络语音识别自适应方法研究

为了解决语音识别中深层神经网络的说话人与环境自适应问题,本文从语音信号中的说话人与环境因素的固有特点出发,提出了使用长时特征的自适应方案：首先基于高斯混合模型,建立说话人-环境联合补偿模型,对说话人与环境参数进行估计,将此参数作为长时特征;然后,将估计出来长时特征与短时特征一起送入深层神经网络,进行训练。Aurora4实验表明,这一方案可以有效地对说话人与环境因素进行分解,并提升自适应效果。     

人脸识别中基于系数相似性的字典学习算法

在现有的基于稀疏表示分类算法的人脸识别中,使用通过稀疏学习得到的精简字典可以提高识别速度和精确度。metaface学习(Metaface Learning,MFL)算法在字典学习过程中没有考虑同类样本稀疏编码系数之间具有相似性的特点。为了利用这一信息来提高字典的区分性,提出了一种基于系数相似性的metaface学习(Coefficient-Simi-larity-based Metaface earning,CS-MFL)算法。CS-MFL算法的学习过程中,在更新稀疏表示系数阶段加入同类训练样本稀疏编码系数相似的约束项。为了求解包含系数相似性约束的新的最优化问题,将目标函数中的两个l2范数约束项进行合并,将原问题转化为典型l₂- l1问题进行求解。在不同的人脸库上进行实验,结果表明,提出的CS-MFL算法能够获得比MFL算法更高的识别率,说明由CS-MFL算法学习得到的字典更高效且更具区分性。