基于快速神经网络架构搜索的鲁棒图像水印网络算法

为解决深度学习在图像水印算法中计算量大且模型冗余的问题，提高图像水印算法在抵抗噪声、旋转和剪裁等攻击时的鲁棒性，提出基于快速神经网络架构搜索（neural architecture search，NAS）的鲁棒图像水印网络算法。通过多项式分布学习快速神经网络架构搜索算法，在预设的搜索空间中搜索最优网络结构，进行图像水印的高效嵌入与鲁棒提取。首先，将子网络中线性连接的全卷积层设置为独立的神经单元结构，并参数化表示结构单元内节点的连接，预先设定结构单元内每个神经元操作的搜索空间；其次，在完成一个批次的数据集训练后，依据神经元操作中的被采样次数和平均损失函数值动态更新概率；最后，重新训练搜索完成的网络。水印网络模型的参数量较原始网络模型缩减了92%以上，大大缩短了模型训练时间。由于搜索得到的网络结构更为紧凑，本文算法具有较高的时间性能和较好的实验效果，在隐藏图像时，对空域信息的依赖比原始网络更少。对改进前后的2个网络进行了大量鲁棒性实验，对比发现，本文算法在CIFAR-10数据集上对抵抗椒盐噪声和旋转、移除像素行（列）等攻击优势显著；在ImageNet数据集上对抵抗椒盐高斯噪声、旋转、中值滤波、高斯滤波、JPEG压缩、裁剪等攻击优势显著，特别是对随机移除行（列）和椒盐噪声有较强的鲁棒性。     

Space-Confined Atomic Clusters Catalyze Superassembly of Silicon Nanodots within Carbon Frameworks for Use in Lithium-Ion Batteries

Incorporating nanoscale Si into a carbon matrix with high dispersity is desirable for the preparation of lithium-ion batteries (LIBs) but remains challenging. A space-confined catalytic strategy is proposed for direct superassembly of Si nanodots within a carbon (Si NDs⊂C) framework by copyrolysis of triphenyltin hydride (TPT) and diphenylsilane (DPS), where Sn atomic clusters created from TPT pyrolysis serve as the catalyst for DPS pyrolysis and Si catalytic growth. The use of Sn atomic cluster catalysts alters the reaction pathway to avoid SiC generation and enable formation of Si NDs with reduced dimensions. A typical Si NDs⊂C framework demonstrates a remarkable comprehensive performance comparable to other Si-based high-performance half LIBs, and higher energy densities compared to commercial full LIBs, as a consequence of the high dispersity of Si NDs with low lithiation stress. Supported by mechanic simulations, this study paves the way for construction of Si/C composites suitable for applications in future energy technologies.     

Sandwich-like Catalyst–Carbon–Catalyst Trilayer Structure as a Compact 2D Host for Highly Stable Lithium–Sulfur Batteries

Herein, we propose the construction of a sandwich-structured host filled with continuous 2D catalysis–conduction interfaces. This MoN-C-MoN trilayer architecture causes the strong conformal adsorption of S/Li₂S_x and its high-efficiency conversion on the two-sided nitride polar surfaces, which are supplied with high-flux electron transfer from the buried carbon interlayer. The 3D self-assembly of these 2D sandwich structures further reinforces the interconnection of conductive and catalytic networks. The maximized exposure of adsorptive/catalytic planes endows the MoN-C@S electrode with excellent cycling stability and high rate performance even under high S loading and low host surface area. The high conductivity of this trilayer texture does not compromise the capacity retention after the S content is increased. Such a job-synergistic mode between catalytic and conductive functions guarantees the homogeneous deposition of S/Li₂S_x, and avoids thick and devitalized accumulation (electrode passivation) even after high-rate and long-term cycling.     

Anti-Markovnikov Hydroamination of Racemic Allylic Alcohols to Access Chiral γ-Amino Alcohols

A ruthenium-catalyzed formal anti-Markovnikov hydroamination of allylic alcohols for the synthesis of chiral γ-amino alcohols is presented. Proceeding via an asymmetric hydrogen-borrowing process, the catalysis allows racemic secondary allylic alcohols to react with various amines, affording enantiomerically enriched chiral γ-amino alcohols with broad substrate scope and excellent enantioselectivities (68 examples, up to >99 % ee).     

桉树杂交种与其亲本的近红外光谱判别

研究桉树控制授粉后目标性状的基因作用方式是探索其基因重组规律的重要内容。常规的数量统计分析精度往往不高,而DNA分析的专业要求高,且费时费力。该研究利用近红外光谱(NIRs)研究不同基因型桉树杂交种、亲本及杂交种与亲本间近红外光谱信息的关系,探索NIRs用于桉树杂交种与其亲本判别的可行性和准确性。以控制授粉的桉树亲本及其杂交F1代材料为对象,每种基因型从各自田间试验分别选取10个单株,采集树冠中上部新鲜健康叶片。用手持式近红外仪Phazir Rx(1624)采集桉树杂交种与其亲本叶片的NIRs信息。每单株选10片完全生理成熟的健康叶片,避开叶脉扫描其正面光谱5次,以50条NIRs信息的均值代表单个叶片的NIRs信息,最终每个基因型获得10条NIRs信息。对原始NIRs采用二阶多项式S.G一阶导数预处理。预处理后的NIRs用于多元统计分析,首先对桉树杂交亲本和子代样本进行主成分分析(PCA),直观展示不同基因型的分类情况。然后运用簇类独立软模式(SIMCA)和偏最小二乘判别分析(PLS-DA)两种有监督的判别模式验证NIRs用于桉树杂交种与其亲本树种的分类判别效果。PCA结果显示,不同的亲本间、杂交种间及杂交种与亲本间样本的主因子得分可以清晰地将各基因型分开。SIMCA模式判别分析中,桉树杂交种样本到亲本PCA模型的样本距离显示,待判别样本能够形成单独的聚类,且能直观反映两者的遗传相似。PLS-DA判别结果显示,桉树杂交亲本的PLS模型能通过预测其杂交子代的响应变量将其与亲本准确分开。结果表明,桉树叶片的NIRs信息可以准确地反映桉树杂交子代遗传信息的传递规律,NIRs判别模型可以准确地将各种基因型予以区分。因此,NIRs信息不仅可用于桉树杂交种和纯种的定性判别,还可以分析桉树基因重组过程中加性遗传效应的大小,从而为桉树遗传基础分析及其育种改良研究提供理论支撑。     

基于冲击光谱测量的氧化铝晶体高压相变诊断技术

氧化铝晶体是一种优良的光学透明窗口材料,更是地球内部的重要组成物质.利用气炮加载结合冲击光谱测量,不仅能够获得其发光特征,并且根据光谱分布特征得到高压结构相变信息.在自主搭建的冲击光谱动态测试平台上,结合多通道辐射高温计以及ICCD瞬态光谱测试技术在40~120 GPa的压力区间,研究了c切向氧化铝晶体的辐射发光效应.在可见光波段400~700 nm区间获得了氧化铝晶体的发光光谱和辐射温度结果,证实了光谱的结构特征和表观温度值与该压力下氧化铝的结构相变存在明显的关联性.     

An approach of dynamic mesh adaptation for simulating 3‐dimensional unsteady moving‐immersed‐boundary flows

In this paper, we present an approach of dynamic mesh adaptation for simulating complex 3‐dimensional incompressible moving‐boundary flows by immersed boundary methods. Tetrahedral meshes are adapted by a hierarchical refining/coarsening algorithm. Regular refinement is accomplished by dividing 1 tetrahedron into 8 subcells, and irregular refinement is only for eliminating the hanging points. Merging the 8 subcells obtained by regular refinement, the mesh is coarsened. With hierarchical refining/coarsening, mesh adaptivity can be achieved by adjusting the mesh only 1 time for each adaptation period. The level difference between 2 neighboring cells never exceeds 1, and the geometrical quality of mesh does not degrade as the level of adaptive mesh increases. A predictor‐corrector scheme is introduced to eliminate the phase lag between adapted mesh and unsteady solution. The error caused by each solution transferring from the old mesh to the new adapted one is small because most of the nodes on the 2 meshes are coincident. An immersed boundary method named local domain‐free discretization is employed to solve the flow equations. Several numerical experiments have been conducted for 3‐dimensional incompressible moving‐boundary flows. By using the present approach, the number of mesh nodes is reduced greatly while the accuracy of solution can be preserved.     

Fabrication of Microporous Metal–Organic Frameworks in Uninterrupted Mesoporous Tunnels: Hierarchical Structure for Efficient Trypsin Immobilization and Stabilization

Hierarchically porous metal–organic frameworks (HP-MOFs) are promising in various applications. Most reported HP-MOFs are prepared based on the generation of mesopores in microporous frameworks, and the formed mesopores are connected by microporous channels, limiting the accessibility of mesopores for bulky molecules. A hierarchical structure is formed by constructing microporous MOFs in uninterrupted mesoporous tunnels. Using the confined space in as-prepared mesoporous silica, highly dispersed metal precursors for MOFs are coated on the internal surface of mesoporous tunnels. Ligand vapor-induced crystallization is employed to enable quantitative formation of MOFs in situ, in which sublimated ligands diffuse into mesoporous tunnels and react with metal precursors. The obtained hierarchically porous composites exhibit record-high adsorption capacity for the bulky molecule trypsin. The thermal and storage stability of trypsin is improved upon immobilization on the composites.     

g-C3N4催化石墨化碳质材料的制备及光催化性能研究

通常采用以氢氧化物作为造孔剂,过渡金属硝酸盐或氯化物作为石墨化催化剂的传统两步法策略制备多孔石墨化碳材料。然而制备过程中多涉及有毒和腐蚀性试剂,且多步骤的过程耗时较长。本文以双氰胺为原料通过热缩聚反应得到g-C₃N₄,采用高铁酸钾为催化剂一步法实现g-C₃N₄的同步碳化-石墨化,并研究其光催化性能。与传统的两步法相比,该方法耗时少、效率高、无污染。与初始的g-C₃N₄材料相比,石墨化g-C₃N₄衍生碳质材料不仅显著改善了可见光的吸收,而且大大增强了光催化活性。研究了不同石墨化温度对g-C₃N₄衍生碳质材料在可见光下降解甲基橙溶液的影响。700 ℃下制备的衍生碳质材料的降解率为12.4 mg/g。光电化学测试结果表明,多孔g-C₃N₄衍生碳质材料的光生载流子密度、电荷分离和光电流(提高了5.4倍)均得到显著提高。因此,该简便、灵活方法为提高g-C₃N₄衍生碳质材料的吸附和光催化性能提供了一种有前景的、高效的途径。