基于深度图及分离池化技术的场景复原及语义分类网络

在机器视觉感知系统中,从不完整的被遮挡的目标对象中鲁棒重建三维场景及其语义信息至关重要.目前常用方法一般将这两个功能分开处理,本文将二者结合,提出了一种基于深度图及分离池化技术的场景复原及语义分类网络,依据深度图中的RGB-D信息,完成对三维目标场景的重建与分类.首先,构建了一种CPU端到GPU端的深度卷积神经网络模型,将从传感器采样的深度图像作为输入,深度学习摄像机投影区域内的上下文目标场景信息,网络的输出为使用改进的截断式带符号距离函数(Truncated signed distance function, TSDF)编码后的体素级语义标注.然后,使用分离池化技术改进卷积神经网络的池化层粒度结构,设计带细粒度池化的语义分类损失函数,用于回馈网络的语义分类重定位.最后,为增强卷积神经网络的深度学习能力,构建了一种带有语义标注的三维目标场景数据集,以此加强本文所提网络的深度学习鲁棒性.实验结果表明,与目前较先进的网络模型对比,本文网络的重建规模扩大了2.1%,所提深度卷积网络对缺失场景的复原效果较好,同时保证了语义分类的精准度.

结合池化技术和特征组的汉语语义角色标注

利用基于统计机器学习方法进行汉语语义角色标注(Semantic Role Labeling,SRL)存在人工抽取特征工作繁琐低效、模型难以捕捉长句上下文语义信息等问题.对此,提出BiLSTM-MaxPool-CRF融合模型进行汉语SRL,同时进行模型性能优化研究.首先,在训练语料中融入词性、论元标记、短语句法等多层级语言学特征;然后,结合AvgPool技术对特征组进行采样选取;最后,通过多组实验结果表明,相比于未采样提取的多特征组,经过池化技术采样提取的多特征能够显著提高模型的性能.

基于城域组网的运营级WLAN组网技术

WLAN作为有线宽带的延伸,也是无线宽带的重要组成部分。近年来,各地政府纷纷提出建设"无线城市",将其作为城市第5项公共基础设施;中国电信股份有限公司也提出了将WLAN建设成为第4张基础网络,这些因素促进了WLAN的大规模建设和发展。分析了基于城域组网的运营级WLAN的建设目标和组网架构,提出了运营级WLAN的组网技术以及WLAN的安全策略部署建议。

移动互联网时代堆叠套餐资源池化计费研究与应用

文章通过对电信运营商传统计费批价方式痛点的分析,提出了一种领先于传统计费平台的新型计费技术,并阐明了这种技术的几大优势和应用场景,有效解决了移动互联网时代电信运营商所面临的计费套餐设置异常复杂问题,提升了电信运营商的运营效率和同行竞争力。

“媒体处理中心”在播出系统中的技术应用研究

本文从媒体处理中心技术处理的基本内容着手分析,对其各项技术予以概念性的讲述,同时对其实际应用的具体分析,本文通过分别列点的形式对其流程进行了详细的阐述。媒体处理中心技术是通过资源池来实现信息配置的,这一智能性较强的实际应用,使我国各电台的台网建设得到了科技支持。

光大银行全栈云异构算力平台建设实践

智能化转型背景下,如何以更小的算力成本承载更多的AI应用成为业界关注的重要课题。本文结合光大银行全栈云异构算力平台建设实践,在深入分析GPU池化技术演进路线及主流技术方案的基础上,总结梳理了全栈云异构算力平台的技术特点与实际成效,并进一步展望了全栈云异构算力平台的未来发展方向。

高显著性的时空金字塔精简描述符算法研究

当传统时空金字塔层数较多时,特征描述符的维数会非常高,使得此类描述符在训练和测试阶段计算效率非常低。此外,在时空金字塔的分层及每层立方体单元的划分中,至今仍然采用手动划分视频,使得视频划分策略没有强的理论依据。鉴于以上缺点,提出一种高显著性的时空金字塔精简描述符算法。形成的新描述符是所有时空金字塔层中每个立方体单元局部特征的权重和,而不是把所有立方体单元局部特征描述符串联起来形成一个巨大的特征描述符,每个立方体单元的权重可以通过偏最小二乘法自动获取,由此产生的视频全局描述符精简并且具有高的显著性。此外通过观测其精细立方体单元的权重,还可以展现出显著性时空金字塔每个立方体单元及每层金字塔的贡献,由此,可以根据权重自动划分视频。采用HMDB51和YouTube两个动作数据库进行实验验证,与时空金字塔描述符和超稀疏编码向量相比,此描述符精简并能在低维度下取得较好的识别效果。

龙岩市教育云服务平台架构设计

探讨龙岩市教育云服务平台的架构设计问题和业务模块架构,提出"3+1"的架构模式以及针对本平台采用的虚拟化技术和资源池化技术,阐述有代表性的业务模块。平台基于明确的IAAS、PAAS和SAAS划分,实现教育资源的高效分配和自由流动,降低广大教育机构IT设施的运维难度和运维成本,以提高龙岩教育信息化的水平。

AS/RS based yard and yard planning

In this paper, we apply the split-platform automated storage/retrieval systems (SP-AS/RSs) (Hu et al., 2005) to store containers in the yard to improve the yard performance and to increase the utilization of the yard space. The layout of an SP-AS/RS based yard is described in detail. To achieve an efficient operation, we present a novel yard space allocation policy called the ‘second-carrier-based allocation policy’, which can help to alleviate the out-of-sequence problem of containers and the congestion of vehicles at the AS/RS racks. Different allocation policies are compared by an integrated container terminal simulation system. The simulation results show that the second-carrier-based policy is very efficient and has the potential to offer high terminal performance.

Expressions for Entropy Production Rate of Fuel Cells

On the basis of a general model of fuel cells, the entropy production rates of a fuel cell system under different conditions are derived by using theories of electrochemistry and thermodynamics. In order to analyze the influence of the irreversible losses existing in an actual fuel cell, the equivalent circuit of the fuel cell is introduced, so that the irreversible factor of the fuel cell may be determined directly as a function of the internal, leak and load resistances. Moreover, the maximum power output and efficiency of the fuel cell are calculated, the optimal operation of the fuel cell is discussed, and the matching condition of the load resistance is determined.

Solid oxide fuel cell:materials,technology and application

Solid oxide fuel cells (SOFCs) offer a clean, pollution-free technology for the electrochemical conversion of chemical energy of hydrocarbon fuels into electricity. Many programs are being initiated in the United States, Europe, Japan and so on. The funding for SOFC development worldwide has risen dramatically and this trend is expected to continue for at least the next decades. These development programs are also investigating wider applications of SOFCs in stationary, residential, transportation and military sectors. Finally, it is summarized the key materials and fabrication processes of SOFC in this paper.

Development of Solid Oxide Fuel Cell Technology at the Institute of Nuclear Energy Research

This paper shows the development of solid oxide fuel cell （SOFC） technology at the Institute of Nuclear Energy Research. In the development, fabrication processes for planar anode-supported-cell （ASC） by conventional methods and metal-supported-cell （MSC） by atmospheric plasma spraying are well established. Procedures and techniques for stacking and cell/stack performance tests are continuously improved to enhance the quality and reliability. Innovative nano-structured catalysts, in which reduced Pt and CeOz particles dispersed onto the A120~ carriers can effectively prevent the migration and coalescence of the metal crystallites, are thermal stable and possess a conversion ratio higher than 95% for reforming of natural gas. A non-premixed after-burner/reformer is designed and fabricated, and it has passed the prerequisite functional tests. Layouts including stacks, components of BOP, power conditioning and control as well as gases and water supply, are designated for a 1-kW SOFC power system. In compliance with system requirements, operating modes, data acquisition, power conditioning, instrumentations, and control logics have been identified and settled. After successive system validation tests, two modules of 18-cell stacks are allocated into the SOFC system. Test results indicate a thermal self-sustaining system on natural gas is achieved with a power output of around 760 watts.

Fuel Cell Technology for Propulsion and Power Generation of Ships： Feasibility Study on Ocean Research Vessel Sagarnidhi

Rising fuel prices, increasing emission levels and impending environmental regulations made shipping industry to find an alternate for internal combustion engine in 21st century. Fuel cell is a sustainable, emerging technology with negligible pollution. More significantly for a research ship, emission levels need to be substantially low to have quality measurements. A feasibility study is carried-out First time in the world, to drive an ice class multi-disciplinary ORV （Oceanography Research Vessel） Sagarnidbi, using hydrogen powered fuel cell. Sagamidhi is equipped with special equipments viz., Deep Sea winch, specially designed cranes for Launching and retrieval of ROV （Remotely Operable Vehicle）, DSMC （Deep Sea Mining Crawler）, Tsunami systems, manned/unmanned submersible and ACS （Autonomous Coring System） and other facilities that support research in Indian, International and Antarctic waters. Beside this, the propulsion system along with DP （Dynamic Positioning）, centralized air conditioning and special equipments require enormous electrical power. The combustion of diesel oil in an engine, that coupled with an alternator generates electrical power required, along with NOx （Nitrous Oxides）, SOx （Sulphur Oxides） and PM （Particulate Matter） emissions. Shipping industry is the fourth largest contributor to air pollution and carbon emissions, particularly in coastal areas, and the growth rate makes the problem even more critical. Stringent international air pollution regulation and increasing fuel price paves the way for an alternative ＂green emission technology＂. Various fuel cells were analyzed with different combination of fuel, electrolyte and electrodes. From the analysis, it has been found that SOFC （Solid Oxide Fuel Cell） is most suitable for the present scenario. A fuel cell designed with hydrogen as fuel, zirconium oxides stabilized with yttrium oxide as electrolyte and zirconium electrodes is used for 1.5 MW power output and 0.5 MW through regenerator. Volume required for storage of hydrogen is in line with volume of fuel and a high standard safety measures were taken using sensors. The present system saves 3000 MT/annum of diesel oil costing 3,000,000 USD approximately.

基于多维度卷积神经网络的人脸识别方法

为提高传统人脸识别的能力,结合CNN卷积神经网络原理,以ResNet18为主干网络,并在ResNet18中引入残差单元,改进激活函数和池化层,通过多维卷积神经网络解决梯度饱和问题。结果表明,改进后的网络在三个数据集上的识别精度都最高,且在消融实验中也表现出良好性能。改进后的网络模型在JAFFE、CK+和RAF-DB数据集上的人脸表情识别准确率依次是97.54%、95.67%和89.66%。通过以上改进,得出本研究的改进方案可行,实现了对图片更多信息的保留。

Optimization of Dimples in Microchannel Heat Sink with Impinging Jets Part A： Mathematical Model and the Influence of Dimple Radius

With increasing heat fluxes caused by electronic components, dimples have attracted wide attention by researchers and have been applied to microchannel heat sink in modem advanced cooling technologies. In this work, the combination of dimples, impinging jets and microchannel heat sink was proposed to improve the heat transfer performance on a cooling surface with a constant heat flux 500 W/cm2. A mathematical model was ad- vanced for numerically analyzing the fluid flow and heat transfer characteristics of a microchannel heat sink with impinging jets and dimples （MHSIJD）, and the velocity distribution, pressure drop, and thermal performance of MI-ISIJD were analyzed by varying the radii of dimples. The results showed that the combination of dimples and MHSIJ can achieve excellent heat transfer performance; for the MHSIJD model in this work, the maximum and average temperatures can be as low as 320 K and 305 K, respectively when mass flow rate is 30 g/s; when dimple radius is larger than 0.195 mm, both the heat transfer coefficient and the overall performance h/AP of MHSIJD are higher than those of MHSIJ.

Optical analysis of a hybrid solar concentrating Photovoltaic /Thermal (CPV/T) system with beam splitting technique

A novel hybrid solar concentrating Photovoltaic/Thermal (CPV/T) system with beam splitting technique is presented. In this system, a beam splitter is used to separate the concentrated solar radiation into two parts: one for the PV power generation and the other for thermal utility. The solar concentrator is a flat Fresnel-type concentrator with glass mirror reflectors. It can concentrate solar radiation onto solar cells with high uniformity, which is beneficial to improving the efficiency of solar cells. The thermal receiver is separated to the solar cells, and therefore, the thermal fluid can be heated to a relatively high temperature and does not affect the performance of solar cells. A dimensionless model was developed for the performance analysis of the concentrating system. The effects of the main parameters on the performance of the concentrator were analyzed. The beam splitter with coating materials Nb2O3 /SiO2 was designed by using the needle optimization technique, which can reflect about 71% of the undesired radiation for silicon cell(1.1m < 3m) to the thermal receiver for thermal utility. The performance of this CPV/T system was also theoretically analyzed.

Two-dimensional iron-porphyrin sheet as a promising catalyst for oxygen reduction reaction: a computational study

The development of non-platinum(Pt) oxygen reduction reaction(ORR) catalysts with high activity and low cost is of great importance for large-scale commercialization of fuel cells. By means of density functional theory(DFT) computations, we theoretically identified that two-dimensional(2D) iron-porphyrin(Fe-Pp) sheet, in which the active Fe sites are distributed regularly and separately, is an appealing candidate. The pristine Fe-Pp sheet exhibits considerably high catalytic activity and four-electron selectivity for ORR. Especially, the adsorption of ORR intermediates on Fe-Pp sheet can be significantly weakened by the addition of axial cyanogen(CN) ligand, resulting in pronouncedly enhanced ORR activity. More interestingly, the d band center of CN attached Fe-Pp(Fe-Pp-CN) sheet can be further tuned by applying the external tensile or compressive strain, leading to an enhancement or suppression of ORR catalytic performance. In particular, under a small biaxial tensile strain of 2%, the ORR activity of Fe-Pp-CN sheet is even higher than that of Pt and reaches to the top of activity volcano. Our studies open new ways to design effective non-Pt ORR catalysts for fuel cell technology.