基于l0正则化的增量低秩特征学习目标跟踪

为了提高生成型目标跟踪算法在遮挡、背景干扰 等复杂条件下的性能,在稀疏编码模型中引入l₀范数正 则化约束,以减少冗余编码信息并改善目标表观重构效果。同时提出一种新的基于非凸近端 加速梯度的快速迭代算法, 解决由此产生的非凸非光滑优化问题。设计了一种增量低秩学习策略,和传统方法需 要将目标观测数据作为 一个整体进行低秩学习不同,本文方法通过l₀正则化稀疏编码能够有效地对目标低秩特 征子空间进行在线学习和更 新。在多个视频序列上的实验表明:基于l₀正则化的增量低秩学习方法能有效提高目标 跟踪算法的准确率和鲁棒性; 和8种优秀的跟踪算法相比,本文算法在中心误差稳健性和重叠率稳健性两个指标上都取得 了最好结果。     

一种基于单像素相机的压缩感知图像重建优化算法

基于压缩感知图像重构和单像素相机成像的基本原理,对单像素成像系统中的图像重建算法进行了改进优化。基于最小范数类优化算法,结合凸优化算法和非凸优化算法各自的优点,设计了一种逼近L₀范数的数学模型,从而实现了从凸优化向非凸优化算法的迭代逼近,即逼近光滑L₀范数算法。该新型算法以更高的效率和更大的概率逼近原始信号全局最优且尽可能稀疏的解。相较于传统压缩感知图像重建的贪婪类算法和最小范数类算法,该算法使压缩感知重建图像的质量和单像素相机的成像效果均得到了有效提升,并通过实验仿真和实际场景的成像实验验证了该优化算法的可行性。     

基于SDBN和BLSTM注意力融合的端到端视听双模态语音识别

提出一种端到端的视听语音识别算法。在该算法中,通过具有瓶颈结构的深度信念网络（deep belief network,DBN）中引入混合的l_1/2范数和l₁范数构建一种稀疏DBN（sparse DBN,SDBN）来提取稀疏瓶颈特征,从而实现对数据的特征降维,然后用双向长短期记忆网络（bidirectional long short-term memory,BLSTM）在时序上对特征进行模态处理,之后利用一种注意力机制将经过模态处理的唇部视觉信息和音频听觉信息进行自动对齐、融合,最后将融合的视听觉信息通过一个附加了Softmax层的BLSTM进行分类识别。实验表明,该算法能有效地识别视听觉信息,在同类算法中有很好的识别率和顽健性。     

基于联合稀疏矩阵恢复的DOA估计算法

基于联合稀疏矩阵恢复的思想,提出一种新的窄带信号DOA 估计算法。算法通过对计算得到的各帧阵列协方差矩阵进行矢量化操作,构造伪数据矩阵;然后构建过完备的阵列方向矩阵字典,形成联合稀疏信号模型;接着利用联合l_2,0 逼近法求出联合稀疏矩阵的优化解,由此得到信号DOA 的估计值。由于二阶统计量的矢量化操作扩展了阵列孔径,算法能够分辨多于阵元数的信号,同时适用于窄带短时平稳或平稳信号,且不需要预先估计信号源数。计算机仿真结果证明了算法的有效性。     

一种混合范数整形分辨率可调的波束形成器

针对实际应用中需要调节分辨率的现象,提出一种混合范数整形分辨率可调的波束形成器。该方法利用信号的稀疏特性与子空间方法使零陷和旁瓣降低的同时,根据l₂范数具有促进稠密解的特性,加入了对主瓣的l₂范数约束,从而增大主瓣的阵列增益强度,使主瓣宽度变窄。仿真结果表明,调节加权系数可以调节波束图的主瓣宽度、旁瓣高度和零陷深度,实现了分辨率可调的目的。也可以根据实际需要通过调节加权系数实现各性能参数间良好折衷。     

一种新的基于压缩感知理论的稀疏信号重构算法

针对基于l1范数优化的稀疏信号重构算法需要的观测样本数较多,本文以lp范数最小化为目标,结合传统的罚函数(PF)优化思想,给出了基于PF的lp范数迭代重构算法,需要的观测样本数大大低于基于l1范数的优化计算需求,并通过数值实验表明该算法对稀疏信号具有较优的重构效果.     

SAR图像目标超分辨的变范数正则化算法

 基于正则化变分的框架,提出了SAR图像目标超分辨的变范数算法.考虑目标在成像区域中的稀疏特性,利用广义高斯分布对目标区域的幅度进行建模,在Bayes估计的框架下,推导了l_p范数约束的正则化变分模型和广义高斯分布形状参数的关系.采用迭代的方法在逐次估计真实图像的过程中,将p的取值与逐次估计结果相关联,逐步估计目标区域分布的形状参数,并修正l_p范数的具体形式,由此得到变范数的正则化模型.该方法克服了通过经验选取p值的局限,以及由观测数据估计p值的误差.仿真和实测SAR图像的处理结果表明了该方法的有效性.     

基于协作稀疏编码的红外小目标检测

为了实现复杂背景下的红外小目标检测, 提出了一种基于协作稀疏编码(CSC)的红外小目标检测算法。首先通过滑动窗口法提取待测 试图像的图像块,并将 其转化为列向量作为超完备字典;然后采用CSC模型计算每一个图像块在超完 备字典中的系数矩 阵以及误差矩阵,其中系数矩阵的L_2,1范数代表图像的背景信息,而误 差矩阵的L_1,2范数代表红外小目标信 息;进而利用ADMM(alternating directional method of multiplier)算法解 算,得到系数矩阵和误差矩阵;最后通 过误差矩阵重建,得到红外小目标的位置。仿真及公开数据实验结果,证实了本文方法的有 效性。     

基于Tchebichef矩与颜色矢量角度的鲁棒图像哈希算法

为了改善哈希算法对旋转等内容修改的鲁棒性,设计了径向Tchebichef矩耦合颜色矢量角度的鲁棒图像哈希算法。引入2D离散小波变换(DWT),对图像的颜色矢量角度实施分解,获取对应的4个子带,将其低频系数作为结构特征。采用径向Tchebichef矩计算预处理图像的Tchebichef矩,提取全局特征。通过组合这2种特征,以形成中间哈希序列。设计加密函数,对中间哈希完成加密,得到目标哈希序列。计算初始目标与待识别图像的哈希序列之间的l₂范数距离,并将其与预设阈值作比较,完成图像内容的真伪判别。测试数据表明：相对于已有的哈希算法而言,所提算法具备更高的鲁棒性,可以对旋转、颜色与缩放等内容修改做出准确识别。     

VOLTE容灾机制研究及优化方案

_{本文首先对VOLTE现网网络结构及容灾机制进行研究分析,发现现存容灾机制中存在的缺点和不确定性,针对其中的问题,针对性地进行了深入研究和分析,创新的提出了1种快速容灾抢通方案,以达到提升VOLTE业务运维能力和用户业务感知的目的。}     

Calculation of energy band gaps in quaternary iii/v alloys

The dielectric theory of electronegativity is applied to the calculation of the compositional dependence of the energy band gap for quaternary III/V alloys of type A_l-xB_xC_1-yD_y and A_1-x-yB_xC_yD. The departure from linearity of E_G versus x and y is taken to be the sum of two terms, the intrinsic or virtual crystal term and the extrinsic term due to effects of aperiodicity which for one type of alloy may occur on both sublattices. Rather than simply treating the quaternary as an average of the bounding ternary systems, as has been common in the past, the intrinsic departure from linearity is calculated by assuming E_h,i,C, and D_av to vary linearly with x and y. The result is a smaller intrinsic deviation from linearity and a much better fit to existing data in the system Ga_1-xIn_xAs_1-y P_y. The calculation is also applied to three systems where no data exist but which are of great interest because of their potential application for the fabrication of lattice matched tandem solar cells: Ga_l-xA1_xAs_l-ySb_y. Ga_1-x-yA1_xIn_yAs, and GaAs_1-x-yP_xSb_y.     

新形势下河北移动全业务支撑策略研究

随着铁塔公司的成立及国家在“宽带中国”“提速降费”等一系列通信领域的重大战略实施,全业务运营越来越成为河北移动保持领先,实现卓越的重要支点。基于此河北移动启动大规模综合业务区建设,由传统传输网围绕基站建设转向全业务支撑。河北移动将综合业务区建设与全业务机房的选取统筹安排,建设综合业务区与全业务机房的联络光缆,将综合业务区光缆网成为承载重要集客数据专线和4G拉远站的载体,合理、有序、迅速实现“一张光缆网”的建设,鼎力支撑全业务及4G的发展。     

Nickel silicidation techniques for strained Si1?xGex, Si1?x?yGexCy, and Si1?yCy alloys material-device applications

A nickel silicide process for Si_1-xGe_x, Si_1-x-yGe_xC_y, and Si_1-yC_y alloy materials compatible with Si technology has been developed. Low-resistivity-phase (12–20 μΘ cm) nickel silicides have been obtained for these alloys with different low sheet-resistance temperature windows. The study shows that thin (15–18 nm) silicide layers with high crystalline quality, smooth silicide surface, and smooth interface between silicide and the underlying material are achievable. The technique could be used to combine the benefits of Ni silicide and Si_1-xGe_x, Si_1-x-yGe_xC_y, and Si_1-yC_y alloys. The technique is promising for Si or Si_1-xGe_x, Si_1-x-yGe_xC_y, and Si_1-yC_y alloy-based metal-oxide semiconductor, field-effect transistors (MOSFETs) or other device applications.     

Low pressure metalorganic chemical vapor deposition of InP and related compounds

The low pressure metalorganic chemical vapor deposition epitaxial growth and characterization of InP, Ga_0.47In_0.53 As and Ga_xIn_1-xAs_yP_1-y, lattice-matched to InP substrate are described. The layers were found to have the same etch pit density (EPD) as the substrate. The best mobility obtained for InP was 5300 cm² V⁻¹S⁻¹ at 300 K and 58 900 cm² V⁻¹ S⁻¹ at 77₂K, and for GaInAs was 11900 cm² V⁻¹ S⁻¹ at 300 K, 54 600 cm² V⁻¹ S⁻¹ at 77 K and 90 000 cm V⁻¹S⁻¹ at 2°K. We report the first successful growth of a GaInAs-InP superlattice and the enhanced mobility of a two dimensional electron gas at a GaInAs -InP heterojunction grown by LP-MO CVD. LP MO CVD material has been used for GaInAsPInP, DH lasers emitting at 1.3 um and 1.5 um. These devices exhibit a low threshold current, a slightly higher than liquid phase epitaxy devices and a high differential quantum efficiency of 60%. Fundamental transverse mode oscillation has been achieved up to a power outpout of 10 mW. Threshold currents as low as 200 mA dc have been measured for devices with a stripe width of 9 um and a cavity length of 300 um for emission at 1.5 um. Values of T in the range 64–80 C have been obtained. Preliminary life testing has been carried out at room temperature on a few laser diodes (λ = 1.5μm). Operation at constant current for severalthousand hours has been achieved with no change in the threshold current.     

Characterization of a rapid thermal anneal TiNxOy/TiSi2 contact barrier

In this paper, the physical and electrical properties of a TiN_xO_y/TiSi₂ dual layer contact barrier are reported. The TiN_xO_y/TiSi₂ barrier was formed by rapidly annealing a Ti thin film on Si in an N₂ ambient. During this process, the Ti film surface reacts with N₂ to form a TiN_xO_y skin layer and the bulk of the Ti film reacts with Si to form an underlying TiSi₂ layer. The influences of rapid thermal anneal (RTA) conditions on the TiN_xO_y layer were investigated by varying the RTA temperature from 600 to 1100° C and cycle duration from 30 to 100 s. It is found that the resulting TiN_xO_y and TiSi₂ layer thicknesses are dependent on RTA temperature and the starting Ti thickness. For a starting Ti thickness of 500Å, 150Å thick TiN_xO_y and 800Å thick TiSi₂ are obtained after an RTA at 900° C for 30 s. The TiN_xO_y thickness is limited by a fast diffusion of Si into Ti to form TiSi₂. When a Ti film is deposited on SiO₂, Ti starts to react with SiO₂ from 600° C and a significant reduction of the SiO₂ thickness is observed after an RTA at 900° C. The resulting layer is composed of a surface TiN_xO_y layer followed by a complex layer of titanium oxide and titanium suicide. In addition, when Ti is depos-ited on TiSi₂, thicker TiN_xO_y and TiSi₂ layers are obtained after RTA. This is because the TiSi₂ layer retards the diffusion of Si from the underlying substrate into the Ti layer. NMOSFETs were fabricated using the TiN_xO_y/TiSi₂ as a contact barrier formed by RTA at 900° C for 30 s and a significant reduction of contact resistance was obtained. In addition, electromigration test at a high current density indicated that a significant improvement in mean time to failure (MTF) has been obtained with the barrier.     

Processing of large YBa2 Cu3Ox domains for levitation applications by a Nd1+x Ba2−x Cu3 Oy-Seeded melt-growth technique

YBa₂Cu₃O_x domains for levitation applications have been produced by a seeding technology that includes Nd_1+x Ba_2−x Cu₃O_y seeds and melt-processing technologies such as conventional melt-textured growth, melt-texturing with PtO₂ and Y₂BaCuO₅ additions, and the new solid-liquid-melt-growth technology. Large domains (∼20 mm) with high levitation forces (F₁ up to 8.2 N) have been produced. The reproducibility of the results is good, and the capability of producing a large number of pellets in a single batch indicates good potential for the production of large amounts of this material.     

Transport properties of thermoelectric materials for coolers

The Seebeck coefficient, thermal conductivity, electrical conductivity and Hall coefficient of cooler grade, p-and n-type ternary alloys of Bi₂Te₃-Sb₂Te₃-Sb₂Se₃ were measured between 10 and 300 K. Between 300 K and about 150 K the temperature dependence of the transport properties can be explained by assuming nondegeneracy and a lattice scattering mechanism. The difference between the temperature dependence of the Hall effect in n-and p-type alloys can be explained by the presence of sub-bands of light and heavy holes in the valence band of p-type alloys.     

The effect of trimethylaluminum concentration on the incorporation of P In AlxGa1−xPyAs1−y grown by organometallic vapor phase epitaxy

The organometallic vapor phase epitaxial (OM-VPE) growth of Al_xGa_1-xP_yAs_1-y on graded GaP_yAs_1-yGaAs in the compositional range 0 < x < 0.9 and 0 < y < 0.6 is reported. It is found that composition control can be easily achieved, and that the vapor phase ratio of trimethylaluminum to trimethylgallium strongly influences the incorporation of P in the solid. A model is developed which explains this in terms of competing reaction rates. The model gives a good fit to the experimental data.     

基于InP衬底的晶格失配In0.68Ga0.32As的MOCVD生长及其特性研究

采用MOCVD生长技术在InP衬底上成功实现了晶格失配的3μm In0.68 Ga0.32As薄膜生长.通过As组分的改变,利用张应变和压应变交替补偿的InAsxP1-x应变缓冲层结构来释放由于晶格失配所产生的应力,在InP衬底上得到了与In0.68Ga0.32 As晶格匹配的InAsxP1-x“虚拟”衬底,通过对缓冲层厚度的优化,使应力能够在“虚拟”衬底上完全豫弛.通过原子力显微镜(AFM)、高分辨XRD、透射电镜(TEM)和光致发光(PL)等测试分析表明,这种释放应力的方法能够有效提高In0.68 Ga0.32 As外延层的晶体质量.