基于SSD的多因素融合的驾驶疲劳检测研究

为了降低因疲劳驾驶而导致的事故发生率,提出一种利用卷积神经网络与人脸特征点、疲劳判定指标相融合的方法,共同构建疲劳驾驶检测模型。首先利用SSD（Single Shot MultiBox Detector）网络定位驾驶员的眼睛与嘴巴区域,VGG16 网络学习这两个区域所包含的疲劳特征;同时再结合人脸68特征点、眼睛纵横比（Eye Aspect Ratio, EAR）和嘴巴纵横比（Mouth Aspect Ratio, MAR）共同判定驾驶疲劳状态。最后,在相同测试集下分别计算SSD算法和Faster-RCNN算法的平均精度均值mAP;在YawDD数据集上应用此模型;并通过模拟驾车环境来验证此模型的可行性。实验结果表明,SSD算法要优于Faster-RCNN算法,并且此模型在YawDD数据集上的检测准确率约达97.2%,摄像头也能对驾驶员的状态进行实时检测。此模型对疲劳状态的检测十分有效,可在一定程度上降低因疲劳驾驶而导致的事故发生率。     

Reduction of Escherichia coli O157:H7 and Salmonella enteritidis on mung bean seeds and sprouts by slightly acidic electrolyzed water

High microbial populations on mung beans and its sprouts are the primary reason of a short shelf life of these products, and potentially present pathogens may cause human illness outbreak. The efficiency for inactivating Escherichia coli O157:H7 (E. coli O157:H7) and Salmonella enteritidis (S. enteritidis), which were artificially inoculated on mung bean seeds and sprouts, by means of slightly acidic electrolyzed water (SAEW, pH 5.0 to 6.5) generated through electrolysis of a mixture of NaCl and hydrochloric acid solution in a non-membrane electrolytic chamber, was evaluated at the different available chlorine concentrations (ACCs, 20–120 mg/l) and treatment time (3–15 min), respectively. The effect of SAEW treatment on the viability of seeds was also determined. Results indicate that the ACC had more significant effect on the bactericidal activity of SAEW for reducing both pathogens on the seeds and sprouts compared to treatment time (P < 0.05). The seeds and sprouts treated with SAEW at ACCs of 20 and 80 mg/l resulted in a reduction of 1.32–1.78 log₁₀ CFU/g and 3.32–4.24 log₁₀ CFU/g for E. coli, while 1.27–1.76 log₁₀ CFU/g and 3.12–4.19 log₁₀ CFU/g for S. enteritidis, respectively. The germination percentage of mung bean seeds was not significantly affected by the treatment of SAEW at an ACC of 20 mg/l for less than 10 min (P > 0.05). The finding of this study implies that SAEW with a near-neutral pH value and low available chlorine is an effective method to reduce foodborne pathogens on seeds and sprouts with less effects on the viability of seeds.     

云计算中的群体智能与不确定性知识表示

The lasting evolution of computing environment, software engineering and interaction methods leads to cloud computing. Cloud computing changes the configuration mode of resources on the Internet and all kinds of resources are virtualized and provided as services. Mass participation and online interaction with social annotations become usual in human daily life. People who own similar interests on the Internet may cluster naturally into scalable and boundless communities and collective intelligence will emerge. Human is taken as an intelligent computing factor, and uncertainty becomes a basic property in cloud computing. Virtualization, soft computing and granular computing will become essential features of cloud computing. Compared with the engineering technological problems of IaaS (Infrastructure as a service), PaaS (Platform as a Service) and SaaS (Software as a Service), collective intelligence and uncertain knowledge representation will be more important frontiers in cloud computing for researchers within the community of intelligence science.     

安全高效的三因素远程身份认证协议

为有效提高远程身份认证协议的安全性能,将椭圆曲线、挑战/应答机制、生物特征、智能卡和口令认证技术相结合,提出了一种安全高效的三因素身份认证协议。协议采用椭圆曲线的点乘运算,既提升了安全性又降低了运算复杂度。认证双方使用挑战/应答机制的3次握手来实现双向认证,协议的认证过程引入随机数,并完成会话密钥协商。协议的安全性能基于Hash函数的单向性、椭圆曲线的离散对数难题和生物特征的安全特性。通过对多种已知攻击的形式化推演,证明该协议能抵御各种常见攻击,具有较高的安全性能。经仿真实验证明,协议具有更高的运算效率。     

塔里木盆地顺北地区中-下奥陶统“断控”缝洞系统划分与形成机制

碳酸盐岩地层普遍发育各类成因机制的非均质性缝洞储层,其孔隙系统的形成演化受岩石的地质演化控制。一般情况下,这类非均质性缝洞储层的形成与地层流体系统中各种不饱和性流体的热化学作用方式和所能带走的物质体积总和有关。但是,塔里木盆地顺北地区发现的“断控”缝洞储层未经历不饱和性流体的溶蚀改造作用,其空间展布明显受到断裂带边界约束,勘探证实可以形成商业价值储层。因此,断裂活动能否发生大规模的物质体积调整,形成完整的缝洞储集系统,需要引起重视。通过应用三维地震、钻井异常工况、测井、元素录井、压恢试井、干扰试井和生产动态资料等,综合表征“断控”储层特征,反演“断控”缝洞单元和储集系统。针对走滑断裂活动过程中的岩石物理特性、构造增容机理和流体改造机制,开展了脆性岩层应变方式、断层封闭性控制因素、水-岩作用方式等研究,提出断裂带内部受岩体错动、破碎及力学-热化学作用而发生的物质体积调整是顺北地区“断控”缝洞储层形成的主导因素,其孔隙系统的形成、演化与断裂活动关系密切。当“断控”缝洞系统形成、演化过程中无法忽视流体影响时,可依据流体性质和作用方式等,细分出“岩溶型”和“热溶型”亚类,作为“断控”储层概念的理论衍生。     

Effect of Signature Sequence on the Performance of an Asynchronous Multiple-Chip-Rate DS/CDMA System

The effects of signature sequence on the performance of an asynchronousmultiple-chip-rate (MCR)-DS/CDMA system are considered.The equations for the multiple access interference (MAI)and the variance of the MAI are developed.The bit error rate is derived by using the Gaussianapproximation.The deterministic signature sequences such as maximal-lengthsequences and Gold sequences are compared with random signature sequencesin terms of the variance of the MAI and bit error rate.We find that the performance of signature sequences in an asynchronousMCR-DS/CDMA system depends not onlyon the crosscorrelation properties of sequences but also on the carrierfrequencyseparation between the interfering and desired users.     

Chromosomal localization of mouse Cenpa gene

Using a previously isolated mouse centromere protein A (Cenpa) probe, we have localized the gene to the proximal region of mouse Chromosome 5, between the known Il6 and Yes1 loci near [Adra2C-D5H4S43-Hdh]. Comparison of this localization with that of human CENPA, which maps to chromosome 2, is consistent with the presence of a new region of conserved synteny between the two species.     

Regularization parameter estimation for feedforward neural networks

Under the framework of the Kullback-Leibler (KL) distance, we show that a particular case of Gaussian probability function for feedforward neural networks (NNs) reduces into the first-order Tikhonov regularizer. The smooth parameter in kernel density estimation plays the role of regularization parameter. Under some approximations, an estimation formula is derived for estimating regularization parameters based on training data sets. The similarity and difference of the obtained results are compared with other work. Experimental results show that the estimation formula works well in sparse and small training sample cases.