基于F-CNN的雷达目标辨识算法

针对雷达真实目标、地杂波和密集假目标的辨识问题，提出了一种基于分解卷积神经网络的雷达目标辨识算法。以深度可分离卷积为基础建立分解卷积神经网络模型。为了减少模型参数，通过减少卷积核数量和全连接层连接节点数量，减少识别特征种类，建立了精简分解卷积神经网络。实测数据的处理结果表明，该算法与现有卷积神经网络方法相比，精简分解卷积神经网络对真实目标样本、地杂波样本和密集假目标样本具有更高的识别正确率，且精简模型参数数量不到现有方法的十分之一。     

多尺度特征融合与极限学习机结合的交通标志识别

针对卷积神经网络只能使用相同尺寸图像和卷积核的网络进行特征提取,导致提取的特征不全面,在交通标志识别中因车载摄像头与交通标志的位置不断变化影响交通标志的识别精度等问题,提出了一种基于多尺度特征融合与极限学习机结合的交通标志识别方法。首先,将预训练适应3种不同尺寸图像的网络模型作为实验的初始模型;然后,融合3个网络模型构建多尺度卷积神经网络,将3个预训练网络的参数级联到融合模型的全连接层,对融合模型的全连接层进行训练,采用随机梯度下降算法更新网络参数;最后,将融合后的模型作为特征提取器提取特征,把提取到的多尺度特征送入极限学习机,实现交通标志识别。实验采用德国交通标志数据库(GTSRB)对算法性能进行测试,实验结果显示,多尺度特征融合与极限学习机结合的网络识别精度为99.23%,识别速度为46ms。相对于预训练的网络,网络的分类精度分别提高了2.35%,3.22%,3.74%。多尺度特征融合能够有效提取交通标志图像的特征信息,极限学习机可以提高分类精度和分类时间,该方法能满足交通标志识别的准确性和实时性的要求。     

数字图像识别的代价函数选择和性能评价

针对传统二次代价函数在卷积神经网络训练过程中图像识别准确率不高的问题,提出基于交叉熵代价函数的卷积神经网络算法。经数学推导,证明交叉熵代价函数较二次代价函数在图像识别中精度更高,并应用MNIST数据集和CIFAR-10数据集,使用AlexNet卷积神经网络,分别采用二次代价函数和交叉熵代价函数对图像识别模型进行训练,当数字图像识别精确率和损失值稳定后,使用测试数据对代价函数进行多次测试,对比识别准确率。仿真结果表明,此方法不仅能提高数字图像识别的准确率,而且相较于传统的代价函数,训练模型速度更快,明显缩减了训练深度神经网络模型的过程。     

改进的全卷积神经网络的脑肿瘤图像分割

针对现有机器学习算法分割脑肿瘤图像精度不高的问题,提出一种基于改进的全卷积神经网络的脑肿瘤图像分割算法。算法首先将FLAIR、T2和T1C三种模态的MR脑肿瘤图像进行灰度归一化,随后利用灰度图像融合技术得到肿瘤信息更加全面的预处理图像;然后采用融合三次脑肿瘤特征信息的改进全卷积神经网络对预处理图像进行粗分割,并且在每个卷积层后加入批量正则化层以加快网络训练的收敛速度,提高训练模型精度;最后融合全连接条件随机场细化粗分割结果中的脑肿瘤边界。实验结果表明,相较于传统的卷积神经网络脑肿瘤图像分割算法,本算法在分割精度和稳定性上有了较大提升,平均Dice可达91.29%,实时性较好,利用训练模型平均1s内可完成单张脑肿瘤图像的分割。      

一种基于深度学习的交通标志识别新算法

针对特定场景交通标志精度低与识别速度慢的问题,基于交通标志边缘信息与卷积神经网络,提出了一种交通标志图像识别T-YOLO算法.该算法基于YOLOv2算法检测思想,融合残差网络、卷积层填充0等结构,下采样舍弃池化层改用卷积层,并提取边缘信息与上采样以提升精度,设计7层特征提取网络以缩短识别速度,随后使用Softmax函数归一化实现多分类,并采用批量归一化、多尺度训练等方法缩短训练时间.实验表明,该算法真实有效,图形处理单元(Graphic Process-ing Unit,GPU)平台上最快检测速度13.69 ms/frame,每帧缩短9.51 ms,最高平均准确率97.3％,提高7.1％,满足实时高精度识别要求.与其他算法相比,该算法在交通标志识别速度与精度方面均有大幅提高,更加适用于现实场景,更贴近车载嵌入式系统.     

基于深度学习的运动车辆与静止车辆识别

本文提出了一种基于YOLOv3算法的运动车辆与静止车辆的识别方法.其利用卷积神经网络提取运动车辆与静止车辆的特征,对网络的输出层Softmax进行修改,并通过大量实验优化权重模型参数;利用AICITY CHALLENGE数据集制作训练和检测数据库.测试结果表明,改进后的YOLOv3算法能更好地识别多种现实生活场景中的运动车辆与静止车辆,取得了95.55％的mAP与34.7 frame/s的检测速度,具有很好的检测性能与实时性;检测精度达到了98％,足够满足实用需求.     

基于多信息流动卷积神经网络的行人再识别

行人再识别问题中,由于视角、光照和行人姿态等因素的变化,导致难以提取有效的行人特征,降低识别精度.而深度神经网络在训练样本较少的情况下较难训练,易出现过拟合现象.针对上述问题,本文提出一种多信息流动卷积神经网络（Multi-information Flow Convolutional Neural Network,MiF-CNN）模型,模型中包含一个特殊的卷积结构,该结构中每层卷积层提取到的特征与后续所有卷积层的输入相连接,增强了网络的特征信息流动性和梯度的反向传播效率,使得模型提取到的行人特征更具判别力.采用多损失函数组合方式训练网络模型,更好的区分行人类别.最后利用欧氏距离对行人特征相似性进行排序.在标准行人再识别数据集VIPeR和CUHK01上的实验表明,本文方法进一步提高了行人再识别精度,并有效改善了深度神经网络的过拟合现象.     

基于BERT-AWC的文本分类方法研究

针对现有文本分类算法处理中文数据时存在的分类精度低、参数量庞大、模型难训练等问题,对BERT算法进行了优化.BERT算法处理中文文本时无法提取词向量特征,为此提出了均匀词向量卷积模块AWC.通过在传统卷积神经网络中引入注意力机制来提取可靠词向量特征,再进一步获取到文本的局部特征,由此弥补了BERT模型无法提取词向量的缺...     

基于卷积神经网络的网络隐写分析方法研究

针对现有网络隐写分析算法特征提取难度大、算法适用范围单一的问题,文章提出了一种基于卷积神经网络的网络隐写分析方法。对网络数据流进行预处理,将所有数据包处理成大小相同的矩阵,最大限度地保留数据特征完整性;使用异构卷积进行特征提取,减少模型计算量及参数数量,加快模型收敛速度;取消池化层,提高模型训练效率。与传统网络隐写分析方法相比,模型能够自动提取数据特征,识别多种网络隐写算法。     

应用深度神经网络和集成学习的电台个体识别

提出了一种基于深度神经网络的个体智能识别方法,可用于电台个体分类识别.该方法构建集成多子网络的一维深度卷积模型,以电台时序信号作为模型输入,进行电台个体分类.利用深度神经网络自动特征化的能力,该方法从时序信号中自动获取个体特征,从而以端到端的形式实现从电台信号识别电台个体.该方法能够免去基于专家知识的特征提取工作,自动提取的个体深度特征还有助于区分传统特征无法区分的高度相似电台个体.实验证明,该方法能有效降低模型调参设计难度,能减轻单一网络带来的特征提取识别过拟合问题,能提高电台个体识别算法的泛化能力与鲁棒性.在信噪比12 dB的条件下,对10类电台8PSK调制信号进行特征提取与识别,整体正确率91.83％,平均正确率为89.12％;对MSK调制信号进行特征提取与识别,平均分类精度为89.1％.     

Curved crystal spectrometer for the measurement of X-ray lines from laser-produced plasmas

In order to diagnose the laser-produced plasmas, a focusing curved crystal spectrometer has been developed for measuring the X-ray lines radiated from a laser-produced plasmas. The design is based on the fact that the ray emitted from a source located at one focus of an ellipse will converge on the other focus by the reflection of the elliptical surface. The focal length and the eccentricity of the ellipse are 1350 mm and 0.9586, respectively. The spectrometer can be used to measure the X- ray lines in the wavelength range of 0.2-0.37 nm, and a LiF crystal （200） （2d = 0.4027 nm） is used as dispersive element covering Bragg angle from 30° to 67.5°. The spectrometer was tested on Shengnang- Ⅱ which can deliver laser energy of 60-80 J/pulse and the laser wavelength is 0.35 μm. Photographs of spectra including the 1 s2p ^1P1-1s^2 ^1S0 resonance line（w）, the 1s2p ^3P2-1s^2 1S0 magnetic quadrupole line（x）, the 1s2p ^3P1-1 s^2 ^1S0 intercombination lines（y）, the 1 s2p ^3S~1-1 s^2 ^1S0 forbidden line（z） in helium-like Ti Ⅹ Ⅺ and the 1 s2s2p ^2P3/2-1 s622s ^2S1/2 line（q） in lithium-like Ti Ⅹ Ⅹhave been recorded with a X-ray CCD camera. The experimental result shows that the wavelength resolution（λ/△ 2） is above 1000 and the elliptical crystal spectrometer is suitable for X-ray spectroscopy.     

Preparation and Characterization of TaN ALD Precursors

High purity organic-tantalum precursors for thin film ALD TaN were synthesized and characterized.Vapor pressure and thermal stability of these precursors were studied.From the vapor pressure analysis,it was found that TBTEMT has a higher vapor pressure than any other published liquid TaN precursor,including TBTDET,TAITMATA,and IPTDET.Thermal stability of the alkyl groups on the precursors was investigated using a 1H NMR technique.The results indicated that the tertbutylimino group is the most stable group on TBTDET and TBTEMT as compared to the dialkylamido groups.Thermal stability of TaN precursors decreased in the following order:TBTDET ＞ PDMAT ＞ TBTEMT.In conclusion,precursor vapor pressure and thermal stability were tuned by making slight variations in the ligand sphere around the metal center.     

Electronic Structure of Semiconductor Nanocrystals

This paper reviews our recent development of the use of the large-scale pseudopotential method to calculate the electronic structure of semiconductor nanocrystals, such as quantum dots and wires, which often contain tens of thousands of atoms. The calculated size-dependent exciton energies and absorption spectra of quantum dots and wires are in good agreement with experiments. We show that the electronic structure of a nanocrystal can be tuned not only by its size,but also by its shape. Finally,we show that defect properties in quantum dots can be significantly different from those in bulk semiconductors.     

Critical Stage Rule-Based Real Time Dispatch（RTD）System in Highly-Mixed-Products （HMP） FAB

An improving utilization and efficiency of critical equipments in semiconductor wafer fabrication facilities are concerned. Semiconductor manufacturing FAB is one of the most complicated and cost sensitive environments. A good dispatching tool will make big difference in equipment utilization and FAB output as a whole. The equipment in this paper is In-Line DUV Scanner. There are many factors impacting utilization and output on this equipment group. In HMP environment one of the issues is changing of reticule in this area and idle counts due to load unbalance between equipments. Here we'll introduce a rule-based RTD system which aiming at decreasing the number of recipe change and idle counts among a group of scanner equipment in a high-mixed-products FAB.     

GaAsSb/InAlAs PA DHBTs Grown by Production MBE

The epi material growth of GaAsSb based DHBTs with InAlAs emitters are investigated using a 4 × 100mm multi-wafer production Riber 49 MBE reactor fully equipped with real-time in-situ sensors including an absorption band edge spectroscope and an optical-based flux monitor. The state-of-the-art hole mobilities are obtained from 100nm thick carbon-doped GaAsSb. A Sb composition variation of less than ± 0.1 atomic percent across a 4 × 100mm platen configuration has been achieved. The large area InAlAs/GaAsSb/InP DHBT device demonstrates excellent DC characteristics,such as BVCEO＞6V and a DC current gain of 45 at 1kA/cm2 for an emitter size of 50μm × 50μm. The devices have a 40nm thick GaAsSb base with p-doping of 4. 5 × 1019cm-3 . Devices with an emitter size of 4μm × 30μm have a current gain variation less than 2% across the fully processed 100mm wafer. ft and fmax are over 50GHz,with a power efficiency of 50% ,which are comparable to standard power GaAs HBT results. These results demonstrate the potential application of GaAsSb/InP DHBT for power amplifiers and the feasibility of multi-wafer MBE for mass production of GaAsSb-based HBTs.     

Influence of polarization extinction ratio on distributed polarization coupling detection

Distributed polarization coupling in polarization-maintaining fibers can be detected by using a white light Michelson interferometer. This technique usually requires that only one polarization mode is excited. However, in practical measurement, the injection polarization direction could not be exactly aligned to one of the principal axes of the PMF, so the influence of the polarization extinction ratio should be considered. Based on the polarization coupling theory, the influence of the incident polarization extinction on the measurement result is evaluated and analyzed, and a method for distributed polarization coupling detection is developed when both two orthogonal eigenmodes are excited.     

Noise character analysis and the restraint of self-pulsation laser diodes

We calculate the Langevin noise sources of self-pulsation laser diodes, analyze the effects of active region noise and saturable-absorption region noise on the power fluctuation as well as period fluctuation, and propose a novel method to restrain the noise effects. A visible SIMULINK model is established to simulate the system, The results indicate that the effects of noise in absorption region can be ignored; that with the increase of DC injecting current, the noise effects enhance power jitter, and nevertheless, the period jitter is decreased; and that with external sinusoidal current modulating the self-pulsation laser diode, the noise-induced power jitter and period jitter can be suppressed greatly. This work is valuable for clock recovery in all-optical network.     

High-quality single-crystal CdSe nanoribbons and their optical properties

Large-scale synthesis of single-crystal CdSe nanoribbons is achieved by a modified thermal evaporation method, in which two-step-thermal-evaporation is used to control CdSe sources＇ evaporation. The synthesized CdSe nanoribbons are usually several micrometers in width, 50 nm in thickness, and tens to several hundred micrometers in length. Studies have shown that high-quality CdSe nanoribbons with regular shapes can be obtained by this method. Room-temperature photolumines-cence indicates that the lasing emission at 710 nm has been observed under optical pumping （266 nm） at power densities of 25-153 kW/cm^2. The full width half maximum （FWHM） of the lasing mode is 0.67 nm     

Propagation of hermite-cosh-gaussian beams passing through ABCD optical system with an annular aperture

By using the expansion of the aperture function into a finte sum of complex Gaussian functions, the corresponding analytical expressions of Hermite-cosh-Gaussian beams passing through annular apertured paraxially and symmetrically optical systems written in terms of ABCD matrix were derived, and they could reduce to the cases with squared aperture. In a similar way, the corresponding analytical expressions of cosh-Gaussian beams through annular apertured ABCD matrix were also given. The method could save more calculation time than that by using the diffraction integral formula directly.     

中国建立EB级存储实验室,致力云计算应用(英文)

正With the support of 863 programs,Sugon Information Industry Co.,Ltd.,set up a dawning EB-class storage laboratory to address massive data storage requirements and largescale cloud computing demonstration applications.The Dawning EB-class cloud storage system adopts advanced fault-tolerant architecture,efficient data fault-tolerant algorithms with user authentication and data encryption policies to deal with the"lost""wrong"and"stolen"problems of data for ensuring the reliability and safety of the EB-class storage system in the public network application environments.The Dawning EB-class Storage Laboratory taking advan-