基于改进残差特征蒸馏的轻量级超分辨率网络

基于深度学习的图像超分辨率算法通常采用递归的方式或参数共享的策略来减少网络参数，这将增加网络的深度，使得运行网络花费大量的时间，从而很难将模型部署到现实生活中。为了解决上述问题，本文设计一种轻量级超分辨率网络，对中间特征的关联性及重要性进行学习，且在重建部分结合高分辨率图像的特征信息。首先，引入层间注意力模块，通过考虑层与层之间的相关性，自适应地分配重要层次特征的权重。其次，使用增强重建模块提取高分辨率图像中更精细的特征信息，以此得到更加清晰的重建图片。通过大量的对比实验表明，本文设计的网络与其他轻量级模型相比，有更小的网络参数量，并且在重建精度和视觉效果上都有一定的提升。     

Superconducting Quantum Metamaterials from High Pressure Melt Infiltration of Metals into Block Copolymer Double Gyroid Derived Ceramic Templates

Mesoscale order can lead to emergent properties including phononic bandgaps or topologically protected states. Block copolymers offer a route to mesoscale periodic architectures, but their use as structure directing agents for metallic materials has not been fully realized. A versatile approach to mesostructured metals via bulk block copolymer self-assembly derived ceramic templates, is demonstrated. Molten indium is infiltrated into mesoporous, double gyroidal silicon nitride templates under high pressure to yield bulk, 3D periodic nanocomposites as free-standing monoliths which exhibit emergent quantum-scale phenomena. Vortices are artificially introduced when double gyroidal indium metal behaves as a type II superconductor, with evidence of strong pinning centers arrayed on the order of the double gyroid lattice size. Sample behavior is reproducible over months, showing high stability. High pressure infiltration of bulk block copolymer self-assembly based ceramic templates is an enabling tool for studying high-quality metals with previously inaccessible architectures, and paves the way for the emerging field of block-copolymer derived quantum metamaterials.     

铝业废阴极炭块资源化利用技术研究

为实现铝业废阴极炭块在钢铁冶金流程中资源化利用，通过深入研究废阴极炭块的相关物性，结合炼钢工艺对炭素及氟化物的物质需求，明确了废阴极炭块的无害化资源利用原理，工业试验结果表明废阴极炭块在炼钢转炉内可较好地实现无害化资源利用。     

导管架平台上部组块双船拆除

为研究复杂海域环境中海洋平台双船拆除方法的安全性与可操作性,采用水动力数值计算与水池模型试验相结合的方法进行研究。将双船水动力耦合计算与水池模型试验进行对比,在相对理想海况下,结合与实际工况接近的规则波作用,进行驳船水动力性能变化与运动响应的幅值分析。在受力分析方面,对甲板耦合装置进行改良设计,对比试验数据与数值计算数据,得到受力在合理范围内的结论。研究内容为实际施工的合理性提出一定的指导性意见。     

Functional organisation of the endomembrane network in the digestive gland of the Venus flytrap: revisiting an old story with a new microscopy toolbox

Up-to-date imaging approaches were used to address the spatiotemporal organisation of the endomembrane system in secretory cells of Dionaea muscipula. Different ‘slice and view’ methodologies were performed on resin-embedded samples to finally achieve a 3D reconstruction of the cell architecture, using ultrastructural tomography, array tomography, serial block face-scanning electron microscopy (SBF-SEM), correlation, and volume rendering at the light microscopy level. Observations of cryo-fixed samples by high-pressure freezing revealed changes of the endomembrane system that occur after trap activation and prey digestion. They provide evidence for an original strategy that adapts the secretory machinery to a specific and unique case of stimulated exocytosis in plant cells. A first secretion peak is part of a rapid response to deliver digestive fluids to the cell surface, which delivers the needed stock of digestive materials ‘on site’. The second peak of activity could then be associated with the reconstruction of the Golgi apparatus (GA), endoplasmic reticulum (ER) and vacuolar machinery, in order to prepare for a subsequent round of prey capture. Tubular continuum between ER and Golgi stacks observed on ZIO-impregnated tissues may correspond to an efficient transfer mechanism for lipids and/or proteins, especially for use in rapidly resetting the molecular GA machinery. The occurrence of one vacuolar continuum may permit continuous adjustment of cell homeostasy. The subcellular features of the secretory cells of Dionaea muscipula outline key innovations in the organisation of plant cell compartmentalisation that are used to cope with specific cell needs such as the full use of the GA as a protein factory, and the ability to create protein reservoirs in the periplasmic space. Shape-derived forces of the pleiomorphic vacuole may act as signals to accompany the sorting and entering flows of the cell.     

碳酸盐岩裂缝性储层常规测井评价方法

哈萨克斯坦K油田石炭系KT-Ⅱ段储层为开阔台地相中低孔、特低渗灰岩储层，裂缝对改善储层的渗流能力至关重要，前人研究的利用常规测井资料评价碳酸盐岩储层裂缝的方法由于受裂缝产状、饱和度、泥浆侵入深浅等多方面因素的影响，存在确定性差、容易误判的缺陷。基于研究区目的层段有限的取心和电成像测井资料，结合孔隙型储层和裂缝-孔隙型储层测井响应特征，提出了利用补偿中子确定基块岩石电阻率与基块声波时差，再比较二者与深侧向电阻率、声波时差的差异特征，进而综合识别储层不同产状裂缝发育段，最后建立了裂缝孔隙度、次生孔隙度、裂缝渗透率、总渗透率等储层参数的测井解释模型，实现了利用常规测井资料对碳酸盐岩油藏储层的裂缝识别及参数定量评价。该方法应用的测井综合评价成果与取心物性分析、生产动态情况能够更好地匹配，为该类油藏的合理高效开发提供了依据。     

Densely connected convolutional network block based autoencoder for panorama map compression

As a novel virtual reality (VR) format, panorama maps are attracting increasing attention, while the compression of panorama images is still a concern. In this paper, a densely connected convolutional network block (dense block) based autoencoder is proposed to compress panorama maps. In the proposed autoencoder, dense blocks are specially designed to reuse feature maps and reduce redundancy of features. Meanwhile, a loss function, which imports a position-dependent weight item for each pixel, is proposed to train and adjust network parameters, in order to make the autoencoder fit to properties of panorama maps. Based on the proposed autoencoder and the weighted loss function, a greedy block-wise training scheme is also designed to avoid gradient vanishing problem and speed up training. During training process, the autoencoder is divided into several sub-nets. After each sub-net is trained separately, the whole network is fine-tuned to achieve the best performance. Experimental results demonstrate that the proposed autoencoder, compared with JPEG, saves up to 79.69 % bit rates, and obtains 7.27dB gain in BD-WS-PSNR or 0.0789 gain in BD-WS-SSIM. The proposed autoencoder also outperforms JPEG 2000, HEVC and VVC in both BD-WS-PSNR and BD-WS-SSIM. Meanwhile, subjective results show that the proposed autoencoder can recover details of panorama images, and reconstruct maps with high visual quality.     

Effects of upholstery materials on the burning behavior of real‐scale residential upholstered furniture mock‐ups

Fire spread and growth on real‐scale four cushion mock‐ups of residential upholstered furniture (RUF) were investigated with the goal of identifying whether changes in five classes of materials (barrier, flexible polyurethane foam, polyester fiber wrap, upholstery fabric, and sewing thread), referred to as factors, resulted in statistically significant changes in burning behavior. A fractional factorial experimental design plus practical considerations yielded a test matrix with 20 material combinations. Experiments were repeated a minimum of two times. Measurements included fire spread rates derived from video recordings and heat release rates (HRRs). A total of 13 experimental parameters (3 based on the videos and 10 on the HRR results), referred to as responses, characterized the measurements. Statistical analyses based on Main Effects Plots (main effects) and Block Plots (main effects and factor interactions) were used. The results showed that three of the factors resulted in statistically significant effects on varying numbers of the 13 responses. The Barrier and Fabric factors had the strongest main effects with roughly comparable magnitudes. Foam was statistically significant for fewer of the responses and its overall strength was weaker than for Barrier and Fabric. No statistically significant main effects were identified for Wrap or Thread. Multiple two‐term interactions between factors were identified as being statistically significant. The Barrier*Fabric interaction resulted in the highest number of and strongest statistically significant effects. The existence of two‐term interactions means that it will be necessary to consider their effects in approaches designed to predict the burning behavior of RUF.     

Using deep learning to combine static and dynamic power analyses of cryptographic circuits

Side-channel attacks have shown to be efficient tools in breaking cryptographic hardware. Many conventional algorithms have been proposed to perform side-channel attacks exploiting the dynamic power leakage. In recent years, with the development of processing technology, static power has emerged as a new potential source for side-channel leakage. Both types of power leakage have their advantages and disadvantages. In this work, we propose to use the deep neural network technique to combine the benefits of both static and dynamic power. This approach replaces the classifier in template attacks with our proposed long short-term memory network schemes. Hence, instead of deriving a specific probability density model for one particular type of power leakage, we gain the ability of combining different leakage sources using a structural algorithm. In this paper, we propose three schemes to combine the static and dynamic power leakage. The performance of these schemes is compared using simulated test circuits designed with a 45-nm library.