基于自适应深度网络的无人机目标跟踪算法

针对无人机（UAV）视频中目标易受到遮挡、形变、复杂背景干扰等问题，提出一种基于自适应深度网络的无人机目标跟踪算法。首先，基于主成分分析（PCA）和卷积神经网络（CNN）算法，设计3阶的自适应深度网络进行目标特征提取，该网络对图像的H、S、I通道分别进行主成分分析学习，将得到的特征向量输入网络进行分层卷积，优化了网络结构，提高了网络的收敛速度和精度。其次，将目标深度特征输入核相关滤波算法进行目标跟踪，通过分析相邻2帧图像的变化率，采用分段自适应调整学习率的算法进行目标模板更新，有效地改善目标遮挡问题。仿真实验结果表明，该算法有效地避免了复杂因素干扰导致的跟踪精度下降，具有较好的鲁棒性，相较于全卷积跟踪（FCNT）算法平均跟踪精度提高了9.62%，平均跟踪成功率提高了11.9%。     

由感知到动作决策一体化的类脑导航技术研究现状与未来发展

随着脑与神经科学以及人工智能技术的持续发展，昆虫和哺乳动物大脑导航机理启发下的感知/认知/路径规划/动作决策一体化类脑导航技术得到了较大发展，可以实现由原始感知信息输入到导航动作决策的直接输出，呈现出接近动物端到端面向目标导航的智能行为，具有提高密集型无人机集群导航鲁棒性、准确性、实时响应动作、自主智能性以及计算效率的潜力。阐述了昆虫和哺乳动物大脑导航机理及其互补对称关系，以及昆虫和哺乳动物大脑导航机理启发的端到端类脑导航技术内涵；论述了类脑导航技术研究进展，包括类脑环境感知、类脑空间认知、面向目标类脑导航；分析了类脑导航向智能化、神经形态系统以及群体导航发展的新趋势；最后讨论了类脑导航技术应用于无人机密集集群系统时存在的挑战。     

Lift coefficient prediction at high angle of attack using recurrent neural network

In this paper, identification of dynamic stall effect of rotor blade is considered. Recurrent Neural Networks have the ability to identify the nonlinear dynamical systems from training data. This paper describes the use of recurrent neural networks for predicting the coefficient of lift (C_Z) at high angle of attack. In our approach, the coefficient of lift (C_Z) obtained from the experimental results (wind tunnel data) at different mean angle of attack θ_mean is used to train the recurrent neural network. Then the recurrent neural network prediction is compared with experimental ONERA OA212 airfoil data. The time and space complexity required to predict C_Z in the proposed method is less and it is easy to incorporate in any commercially available rotor code.     

区域运输规划模型系统

提出一种区域运输规划模型系统,以求从定量的角度,预测一个区域内的货物运输量,继而确定这些运输量对该区域内存在的多种运输方式的偏好结构,最后仿真这些运量在相应的各种运输网络上的分布情况,同时仿真模型还对运输网线路经过新建或扩建后所形成的新的运量分布进行模拟。根据运量分布情况,就可提出初步的规划建设方案。     

平方阻尼在非线性缓冲系统中的作用

本文讨论平方阻尼(以参数a表征)在图(1)所示的缓冲系统中的作用,结果表明:平方阻尼对于降低相对位移峰值d_m总有良好的效果,但不一定降低加速度峰值a_m。在一定速度冲击V下,系统的平方阻尼有两种最佳值,使a_m最小的a_(am)和使缓冲效率γ最高的a_(cm),而平方阻尼最佳条件依赖于系统的缓冲弹簧的刚度变化类型: (1)线性弹簧情况:a_(am)=a_(cm),最佳条件为系统在首程中均匀减速; (2)渐硬弹簧情况:a_(am)=a_(cm),最佳条件为系统在首程中初始和终末加速度相等即; aV~2=ω_0~2q(dm) (3)渐软弹簧情况:a_(cm)>a_(am)。 本文的基础是所谓“首程显峰”,假定即系统在速度冲击V下的响应峰值d_m和a_m出现在第一行程中。当平方阻尼值不太小,缓冲弹簧刚度特性接近反对称时,该假定成立。     

金属多层膜超晶格结构研究

用高分辨的三晶体x—射线衍射仪测量了Zr、Cu金属交替的多层膜的x—射线小角反射谱。用扩展了的Born—wolf光学模型对Zr、Cu多层膜的x—射线小角反射数据进行了定量分析。结果表明，该金属多层膜在其生长方向具有超晶格结构。在两层金属膜之间的界面上，由于内扩散，形成了界面交混层，它使x—射线小角反射所产生的高阶Bragg峰的强度减弱。由于两种金属膜的晶格不匹配，使高阶Bragg峰加宽。界面的粗糙度可用来模拟x—射线小角反射曲线的阻尼振荡效应。     

Specific grinding energy and surface roughness of nanoparticle jet minimum quantity lubrication in grinding

Nanoparticles with the anti-wear and friction reducing features were applied as cooling lubricant in the grinding fluid. Dry grinding, flood grinding, minimal quantity of lubrication（MQL）, and nanoparticle jet MQL were used in the grinding experiments. The specific grinding energy of dry grinding, flood grinding and MQL were 84, 29.8, 45.5 J/mm³, respectively. The specific grinding energy significantly decreased to 32.7 J/mm³ in nanoparticle MQL. Compared with dry grinding, the surface roughness values of flood grinding, MQL, and nanoparticle jet MQL were significantly reduced with the surface topography profile values reduced by 11%, 2.5%, and 10%,respectively, and the ten point height of microcosmic unflatness values reduced by 1.5%, 0.5%,and 1.3%, respectively. These results verified the satisfactory lubrication effects of nanoparticle MQL. MoS₂, carbon nanotube（CNT）, and ZrO₂ nanoparticles were also added in the grinding fluid of nanoparticle jet MQL to analyze their grinding surface lubrication effects. The specific grinding energy of MoS₂ nanoparticle was only 32.7 J/mm³, which was 8.22% and 10.39% lower than those of the other two nanoparticles. Moreover, the surface roughness of workpiece was also smaller with MoS₂ nanoparticle, which indicated its remarkable lubrication effects. Furthermore,the role of MoS₂ particles in the grinding surface lubrication at different nanoparticle volume concentrations was analyzed. MoS₂ volume concentrations of 1%, 2%, and 3% were used.Experimental results revealed that the specific grinding energy and the workpiece surface roughness initially increased and then decreased as MoS₂ nanoparticle volume concentration increased.Satisfactory grinding surface lubrication effects were obtained with 2% MoS₂ nanoparticle volume concentration.     

复合材料层合结构低速冲击后的渗漏性能

对复合材料贮箱结构进行低速冲击试验,并测试冲击后贮箱结构的渗漏性能。试验结果表明,当层板内部的基体损伤和分层损伤构成贯穿的通路后即导致贮箱渗漏,并且贮箱结构的冲击能量渗漏门槛值介于23~25J之间。在试验基础上建立了有限元模型以分析冲击后贮箱结构的内部损伤情况,并判断结构是否渗漏,计算结果与试验吻合良好。在计算模型的基础上对曲面贮箱结构的冲击渗漏性能进行分析。     

含有不平衡飞轮的卫星精确动力学模型研究

针对含不平衡飞轮的卫星动力学模型,给出了不平衡飞轮的转动惯量矩阵。用多刚体建模方法,建立了综合的完整姿态动力学分析模型,提出了质心位置的计算方法。通过飞轮系统的完整精确模型,将飞轮扰动对三轴姿态稳定卫星的影响反映至滚转一俯仰一偏航三轴姿态。Matlab／Simulink环境中仿真验证了该系统角动量为常值。分析了不平衡飞轮对卫星姿态的影响,并解释了模型的应用。研究表明飞轮扰动影响不可忽略,精确的动力学模型对未来航天器的应用有重要意义。     

混合三元系超大集

对于三类混合三元系超大集Ti-OLGD（v）（i=1,2,3）,T1-OLGD（v）和T2-OLGD（v）的存在谱已经完全解决了,而对于T3-OLGD（v）的存在性问题,所知结果很少。本文利用代数和组合设计相结合的方法,主要利用有限域的相关知识,对于T3-OLGD（q-1）（q为素数幂）,给出了一种新的构造方法,从而得到了一系列的结果。