非下采样变换的红外与可见光图像融合

基于非下采样Contourlet变换（NSCT）,提出了一种红外和可见光图像融合算法。针对低频子带系数和各带通方向子带系数分别提出了基于图像物理特征的系数加权选择方式与基于区域能量匹配的系数选择方式,即低频基于区域梯度信息、高频基于区域特征因子的加权与选择结合的图像融合算法。实验结果表明：非下采样Contourlet变换具有较快的运算速度,且经非下采样变换后能量更加集中,可提供更多的图像信息。相对于基于像素的图像融合算法,本文的图像融合算法具有更高的融合性能,是一种更适合图像融合的多尺度几何分析（MGA）工具。     

Design and simulation of sensor fusion using symbolic engines

Sensor fusion is the art of estimating accurate information from noisy multi-sensor data. Due to the complexity of stochastic sensor errors, design and testing of sensor fusion algorithms have been always challenging. Existing design approaches are mainly mission specific with fixed system models that do not verify if the filter can estimate hidden errors. To address these challenges, this paper presents a flexible design and simulation environment for sensor fusion. The environment utilizes symbolic engine as a flexible representation of system models to enable flexible and accurate generation of linearized error models. Inverse kinematic is used to generate pseudo-error-free inertial data to test the ability of the filte to estimate sensor errors. The developed environment is demonstrated on an Attitude and Heading Reference System using Extended Kalman Filter. The demonstration includes both simulation and experimental tests. The designed filter supports both loosely and tightly coupled filtering approaches.     

语音情感识别中的特征选择方法

为了解决传统卷积神经网络在识别中文语音时预测错误率较高、泛化性能弱的问题,首先以深度卷积神经网络(DCNN)-连接时序分类(CTC)为研究对象,深入分析了不同卷积层、池化层以及全连接层的组合对其性能的影响;其次,在上述模型的基础上,提出了多路卷积神经网络(MCNN)-连接时序分类(CTC),并联合SENet提出了深度SE-MCNN-CTC声学模型,该模型融合了MCNN与SENet的优势,既能加强卷积神经网络的深层信息的传递、避免梯度问题,又可以对提取的特征图进行自适应重标定。最终实验结果表明:SE-MCNN-CTC相较于DCNN-CTC错误率相对降低13.51%,模型最终的错误率达22.21%;算法改进后的声学模型可以有效地提升泛化性能。     

Easy Production of “Difficult Peptides” Using Cell-Free Protein Synthesis and a New Methionine Analogue as a Latent Peptide Cleavage Site

Aliphatic γ-chloro-α-amino acids incorporated in place of their canonical analogues through cell-free protein synthesis act as heat-labile linkers, offering a useful strategy for the straightforward production of target peptides as fusion proteins, from which the targets are readily released. Until now, the natural abundance of aliphatic amino acids in peptides has limited the scope of the method, as it leads to undesired cleavage sites in synthesized products, but here the authors report the development of a new cleavable chloro amino acid that incorporates in place of the relatively rare amino acid methionine, thus greatly expanding the scope of producible targets. This new strategy is employed for simplified peptide synthesis with a methionine-free fusion partner, allowing single-site incorporation of the cleavable linker for clean release and easy purification of the target peptide. Its utility is demonstrated through the straightforward preparation of two peptides reported to be challenging targets and not accessible through standard solid-phase chemical methodologies, as well as analogues.     

Controlled-Atmosphere Flame Fusion Single-Crystal Growth of Non-Noble fcc,hcp, and bcc Metals Using Copper,Cobalt, and Iron

The growth of noble-metal single crystals via the flame fusion method was developed in the 1980s. Since then, there have been no major advancements to the technique until the recent development of the controlled-atmosphere flame fusion (CAFF) method to grow non-noble Ni single crystals. Herein, we demonstrate the generality of this method with the first preparation of fcc Cu as well as the first hcp and bcc single crystals of Co and Fe, respectively. The high quality of the single crystals was verified using scanning electron microscopy and Laue X-ray backscattering. Based on Wulff constructions, the equilibrium shapes of the single-crystal particles were studied, confirming the symmetry of the fcc, hcp, and bcc single-crystal lattices. The low cost of the CAFF method makes all kinds of high-quality non-noble single crystals independent of their lattice accessible for use in electrocatalysis, electrochemistry, surface science, and materials science.     

基于YOLOv3的正下无人机视角挖掘机实时检测方法

针对无人机巡检的智能化要求, 提出一种应对高空巡检场景下的实时挖掘机检测模型. 该模型以YOLOv3为基础, 将骨干网络精简至43层, 通过特征融合策略使检测任务在两个尺度上进行. 此外模型还借鉴了focal loss的思想设计损失函数. 文中实验对象为正下无人机视角的挖掘机目标. 在完成了数据集的搭建工作后, 根据正下无人机视角的目标特性进行训练, 使模型达到最优解. 最终经实验验证, 在相同输入尺寸的情况下, 本文所提出的检测模型比YOLOv3准确率更高、鲁棒性更好, 且帧数可提升10帧●s^-1.     

人类首次实现聚变“点火”,激光聚变取得历史性突破

2022年12月13日，美国能源部(DOE)及其下属的国家核安全管理局(NNSA)宣布，劳伦斯利弗莫尔国家实验室(LLNL)利用其建立的国家“点火”装置(NIF),在人类历史上首次实现了聚变产能大于驱动聚变发生的激光能量这一“点火”里程碑，将为美国核武器物理规律和效应研究、核武器库存管理等提供重要支撑，为未来清洁能源的发展铺平新的道路，并为高能量密度物理研究提供新的手段和平台。本文专访了中国工程物理研究院激光聚变研究中心郑万国研究员，就发布会传递信息、惯性约束聚变(ICF)实现途径及存在难点、激光聚变“点火”历程、未来ICF和惯性聚变能(IFE)发展前景，以及激光晶体在ICF和IFE中重要作用等业界广泛关心的几个问题进行解读，以期为读者提供专业的信息，使大家进一步了解ICF发展趋势和IFE发展前景，并针对相关晶体材料开展基础研究及关键技术攻关，牵引和支撑未来激光聚变驱动装置建设。     

基于红外和可见光图像逐级自适应融合的场景深度估计

从图像中恢复场景的深度是计算机视觉领域中的一个关键问题。考虑到单一类型图像在深度估计中受场景不同光照的限制，提出了基于红外和可见光图像逐级自适应融合的场景深度估计方法（PF-CNN）。该方法包括双流滤波器部分耦合网络、自适应多模态特征融合网络以及自适应逐级特征融合网络。在双流卷积中红外和可见光图像的滤波器部分耦合使两者特征得到增强；自适应多模态特征融合网络学习红外和可见光图像的残差特征并将两者自适应加权融合，充分利用两者的互补信息；逐级特征融合网络学习多层融合特征的结合，充分利用不同卷积层的不同特征。实验结果表明:PF-CNN在测试集上获得了较好的效果，将阈值指标提高了5%，明显优于其他方法。     

基于多源异构数据的甲骨学知识图谱构建方法研究

为解决和缓解甲骨学研究难度大、周期长、知识关联性弱、知识共享程度低等问题。基于多源异构数据源，融合基于文献计量学的科学知识图谱(MKD)和基于知识库的知识图谱(KG)，构建了甲骨学融合知识图谱。通过融合两类知识图谱，并基于知识推理进行语义扩展，形成最终的甲骨学知识图谱。其中包含实体148 305个，关系434 032条，可满足甲骨学研究的基本需求。融合MKD和KG两类知识图谱，优势互补，实现甲骨学知识图谱构建，可为其他古籍类知识图谱构建提供借鉴。