基于无标签视频数据的深度预测学习方法综述

基于视频数据的深度预测学习(以下简称"深度预测学习")属于深度学习、计算机视觉和强化学习的交叉融合研究方向,是气象预报、自动驾驶、机器人视觉控制等场景下智能预测与决策系统的关键组成部分,在近年来成为机器学习的热点研究领域.深度预测学习遵从自监督学习范式,从无标签的视频数据中挖掘自身的监督信息,学习其潜在的时空模式表达....     

分数阶MAC系统同步及其在保密通信中的应用

目前,混沌同步控制在保密通信中的应用仍是研究热点.文章对分数阶MAC混沌系统进行数值计算,利用Matlab仿真计算出系统的混沌吸引子,并计算出系统随阶数变化的分岔图.利用李雅普诺夫直接法,选取适合系统的控制器进行驱动控制,从而使系统达到平稳的同步.基于混沌掩盖法,将消息信号进行掩盖处理,并根据线性稳定性原理,利用混沌同...     

高速移动通信系统中OTFS分数多普勒信道估计加窗研究

针对正交时频空间(OTFS)调制系统中分数多普勒信道对应的物理路径信道状态信息估计困难及计算复杂度较高等问题,该文提出一种节省导频资源的脉冲匹配滤波(PRS-PMF)信道估计算法。该算法首先使用数据与导频联合成帧的嵌入式辅助导频方法获得等效信道的估计,然后通过互相关匹配滤波估计出各路径信道状态信息,相比于传统的脉冲导频互相关匹配滤波信道估计算法,能够在降低计算复杂度的同时减少导频资源的占用。在此基础上,对OTFS系统加窗,减少窗口响应主瓣的整数样点数量并降低旁瓣电平,有效改善了等效信道多普勒响应函数的自相关特性,从而降低了其他符号及噪声对估计符号的干扰。     

一种人与人和机器到机器共存下能效最大化的上行用户分配算法

针对5G超密集异构蜂窝网络中人与人(H2H)和机器到机器(M2M)共存场景下有服务质量(QoS)保障和负载均衡的上行用户分配问题,该文提出一种基于匹配理论的用户分配算法.该算法在用户分配过程中同时考虑不同类型节点的接入控制策略,在最大化网络能效的同时,实现节点QoS保障.仿真结果表明,与其他算法相比,该算法不仅能够在能效、负载均衡以及QoS保障方面获得更好的性能,并且能获得与穷举法相近的性能.此外,所提算法的收敛速度很快且不受节点和基站数目变化的影响,适合解决H2H和M2M共存场景中的用户分配问题.     

分数阶混沌系统自适应有限时间追踪控制

基于实现一类参数未知三维分数阶混沌系统有限时间追踪控制的目的,采用分数阶系统稳定性地理对分数阶Zhang混沌系统进行了动力学性质分析,采用有限时间控制法设计了一个自适应有限时间追踪控制器,通过理论计算和MATLAB数值仿真,验证了该控制器的有效性,实现了参数未知分数阶混沌系统的自适应有限时间追踪控制;在保持系统各个参数不变的情况下加入外部随机扰动,通过MATLAB数值仿真可知,该控制器具有抗干扰性。     

温度补偿的延迟线型SAW应变传感器

声表面波(SAW)延迟线作为传感器可对传感器进行编码,从而实现多点分布式测量,因而受到广泛研究。该文采用128°Y-X切向LiNbO₃作为压电基底,制备了不同对数的叉指换能器(IDT)的SAW延迟线,并研究了其应变特性。研究结果表明,IDT对数为10对、20对、30对时,SAW延迟线应变灵敏度分别为1.727 5 ps/με、2.046 7 ps/με、3.256 6 ps/με。SAW延迟线的应变特性和应变方向与声波传播方向间的夹角有关,夹角为0°时,延迟时间随应变的增大而增大;夹角为90°时,延迟时间随应变的增大而减小。利用夹角分别为0°和90°的SAW延迟线构成了差分结构的应变传感器。测试结果表明,该差分结构可抵消温度的影响,即温度变化时也可获得准确的应变。该文研究的差分结构延迟线有望在温度波动环境中测试应变时得到应用。     

一种实时多尺度目标检测识别算法

无人机目标检测与识别任务中,目标随着飞行高度的改变尺寸发生显著变化。常规目标检测模型中,获取的小目标细节信息有限,检测精度较低;而适用于小目标的实时检测模型往往容易丢失大目标的背景信息,降低大目标的检测精度。针对以上多尺度目标检测识别任务难点,提出一种基于改进特征金字塔网络(Feature Pyramid Network, FPN)结构的实时多尺度目标检测识别模型。该模型通过增加特征金字塔层级覆盖更广的目标尺度,获取更为丰富的目标信息;同时,利用跨连接增加不同尺度特征融合的多样性,降低特征传导距离,保留更加完整的尺度特征来提高模型检测识别多尺度目标的性能。通过实验发现,相比于原始网络结构和相同特征层级的四层特征金字塔结构,加入改进特征金字塔结构的多尺度目标检测模型识别性能得到了提升。     

HPRF PD 末制导雷达抗遮挡方法设计

本文介绍了距离遮挡产生的原因,分析了影响距离遮挡现象的主要因素,指出高脉冲重复频率脉冲多普勒（HPRFPD）末制导雷达距离遮挡现象影响严重,因此必须采取抗遮挡措施。介绍了现阶段抗距离遮挡的主要方法,通过分析每种方法的不足之处,指出有效对抗距离遮挡现象的关键在于准确测得目标距离,但是在高脉冲重复频率情况下,精确测距只能通过解模糊来获得。本文比较了目前解模糊的主要方法,选取出适合应用在末制导雷达环境下,解模糊时间短,准确度高的方法。在此基础上设计了末制导雷达抗遮挡策略,分别针对无精确距离信息和有精确距离信息两种情况,设计出不同的抗遮挡策略,通过仿真证明,此方法抗遮挡效果较好。     

Silica‐Coated Manganese Oxide Nanoparticles as a Platform for Targeted Magnetic Resonance and Fluorescence Imaging of Cancer Cells

Monodisperse silica‐coated manganese oxide nanoparticles (NPs) with a diameter of ～35 nm are synthesized and are aminated through silanization. The amine‐functionalized core–shell NPs enable the covalent conjugation of a fluorescent dye, Rhodamine B isothiocyanate (RBITC), and folate (FA) onto their surface. The formed Mn₃O₄@SiO₂(RBITC)–FA core–shell nanocomposites are water‐dispersible, stable, and biocompatible when the Mn concentration is below 50 µg mL^?1 as confirmed by a cytotoxicity assay. Relaxivity measurements show that the core–shell NPs have a T₁ relaxivity (r₁) of 0.50 mM ^?1 s^?1 on the 0.5 T scanner and 0.47 mM ^?1 s^?1 on the 3.0 T scanner, suggesting the possibility of using the particles as a T₁ contrast agent. Combined flow cytometry, confocal microscopy, and magnetic resonance imaging studies show that the Mn₃O₄@SiO₂(RBITC)–FA nanocomposites can specifically target cancer cells overexpressing FA receptors (FARs). Findings from this study suggest that the silica‐coated Mn₃O₄ core–shell NPs could be used as a platform for bimodal imaging (both magnetic resonance and fluorescence) in various biological systems.     

Single‐Crystalline Dodecahedral and Octodecahedralα‐Fe2O3 Particles Synthesized by a Fluoride Anion–Assisted Hydrothermal Method

Despite significant advances in iron oxide nanoparticles, it is still a challenge to synthesize regular polyhedral single‐crystalline α‐Fe₂O₃ particles because the surface energies of several low‐index planes are fairly similar. In the work presented here, well‐dispersed and single‐crystalline dodecahedral and octodecahedral α‐Fe₂O₃ particles are synthesized by a facile hydrothermal method with the aid of F^? anions. The crystalline structure of the polyhedral particles is disclosed by various characterization techniques. The dodecahedral particles are of hexagonal bipyramidal shape and enclosed by twelve equivalent (101) planes. The octodecahedral particles are formed by adding six equivalent (111) planes on the two tips of a dodecahedral particle, that is, they are enclosed by twelve (101) planes and six (111) planes. The existence of F^? anions plays a crucial role in the control of polyhedral particle shape. The function of F^? anions in the shape formation of the polyhedral particles is proposed as follows: 1) A high concentration of exposed Fe³⁺ cations induces preferential adsorption of F^? anions on the (100) plane and leads to the slowest growth along the [100] direction. When the concentration of F^? anions is higher than 24 mM , a stable speed ratio of growth along the [001] and [100] directions results in the exposure of (101) planes. 2) With a lower concentration of F^? anions, six symmetrical (111) planes with low concentration of exposed Fe³⁺ cations are present at the tops of a dodecahedral particle to form an octodecahedron. Furthermore, the dodecahedral and octodecahedral α‐Fe₂O₃ particles show much stronger magnetism than the previously reported α‐Fe₂O₃ nanostructures, having coercivities of 4986 Oe and 6512 Oe, respectively. Such high coercivities are attributed to a large local magnetic anisotropy, which might be induced by the polyhedron with equivalent crystallographic planes and/or the presence of F^? anions.