基于多通道的ICU脑血管疾病死亡风险预测模型

死亡风险预测指根据病人临床体征监测数据来预测未来一段时间的死亡风险。对于ICU病患，通过死亡风险预测可以有针对性地对病人做出临床诊断，以及合理安排有限的医疗资源。基于临床使用的MEWS和Glasgow昏迷评分量表，针对ICU病人临床监测的17项生理参数，提出一种基于多通道的ICU脑血管疾病死亡风险预测模型。引入多通道概念应用于BiLSTM模型，用于突出每个生理参数对死亡风险预测的作用。采用Attention机制用于提高模型预测精度。实验数据来自MIMIC [Ⅲ]数据库，从中提取3?080位脑血管疾病患者的16?260条记录用于此次研究，除了六组超参数实验之外，将所提模型与LSTM、Multichannel-BiLSTM、逻辑回归（logistic regression）和支持向量机（support vector machine, SVM）四种模型进行了对比分析，准确率Accuracy、灵敏度Sensitive、特异性Specificity、AUC-ROC和AUC-PRC作为评价指标，实验结果表明，所提模型性能优于其他模型，AUC值达到94.3%。     

Deacetylated Cellulose Acetate/Polyurethane Composite Nanofiber Membranes with Highly Efficient Oil/Water Separation and Excellent Mechanical Properties

Nowadays, oil pollution has become more serious, which causes great threats both to the ecological environment and human life. In this study, a novel type of multifunctional deacetylated cellulose acetate/polyurethane (d-M_CA:M_TPU) composite nanofiber membranes for oil/water separation are successfully fabricated by electrospinning, which show super-amphiphilicity in air, super-hydrophilicity in oil, and oleophobicity in water. All the d-M_CA:M_TPU composite nanofiber membranes with different mass ratios can be used as water-removing, oil-removing, and emulsion separation substance only by gravity driving force. The highest separation flux for water and oil reaches up to 37 000 and 74 000 L m⁻² h⁻¹, respectively, and all the separation efficiencies are more than 99%. They have outstanding comprehensive mechanics performance, which can be controlled by simply adjusting the mass ratios. They show excellent antifouling and self-cleaning ability, endowing powerful cyclic stability and reusability. Those results show that d-M_CA:M_TPU composite nanofiber membranes have great application prospects in oil/water separation.     

多先验融合的图像显著性目标检测算法

为了更加准确地检测出图像中的显著性目标，提出了多先验融合的显著性目标检测算法。针对传统中心先验对偏离图像中心的显著性目标会出现检测失效的情况，提出在多颜色空间下求显著性目标的最小凸包交集来确定目标的大致位置，以凸包区域中心计算中心先验。同时通过融合策略将凸包区域中心先验、颜色对比先验和背景先验融合并集成到特征矩阵中。最后通过低秩矩阵恢复模型生成结果显著图。在公开数据集MSRA1000和ESSCD上的仿真实验结果表明，MPLRR能够得到清晰高亮的显著性目标视觉效果图，同时F，AUC，MAE等评价指标也比现有的许多方法有明显提升。     

不同硬度PTMG型CPUE与硫化橡胶的性能对比研究

通过改变硬段含量、扩链系数等制备了几种不同硬度的四氢呋喃均聚醚(PTMG)浇注型聚氨酯弹性体(CPUE),讨论了PTMG型CPUE与硫化橡胶(VR)的物理机械性能、拉伸应力应变行为以及动态力学性能的区别。结果表明,随CPUE硬度的提高,材料的刚性增强,力学性能是低硬度VR的数倍。通过对CPUE硬度的调节,可以使之兼具聚氨酯和橡胶的优良性能。在玻璃态温度下CPUE的储能模量随硬度的减小而降低,但都高于VR,而且CPUE的硬度越大,二者相差越大,温度越低,相差也越大;在室温以上温度范围内,CPUE具有比VR更小的力学损耗。     

聚酯类增塑剂的合成与应用研究进展

综述了国内外聚酯增塑剂的合成及应用的最新研究进展，重点介绍了超支化、生物基、石油基聚酯增塑剂的合成与制备方法以及不同类型聚酯增塑剂在聚氯乙烯(PVC)、淀粉、聚乳酸、聚氨酯弹性体与橡胶制品中的最新应用研究进展；并对聚酯增塑剂未来发展方向做了展望。     

（S）-噻吗洛尔半水合物的合成及表征

为了开发β受体阻断剂新药（S）-噻吗洛尔半水合物，采用3-吗啉-4-氯-1，2，5-噻二唑为起始原料，经水解反应得到中间体1（3-吗啉-4-羟基-1，2，5-噻二唑）。中间体1与R-环氧氯丙烷发生醚化反应，经后处理及重结晶得到中间体2 {（R）-4-［4-（环氧乙烷-2-基甲氧基）-1，2，5-噻二唑-3-基］吗啉}。中间体2经胺化反应、马来酸成盐及重结晶得到（S）-马来酸噻吗洛尔。（S）-马来酸噻吗洛尔经游离、纯水转晶得到符合药典标准的（S）-噻吗洛尔半水合物，总收率14.05%且e.e.值为99.66%。最终成品经IR、1H-NMR、13C-NMR、MS、TGA、DSC表征，并优化各步反应条件。结果表明:以三乙胺为醚化反应缚酸剂75 ℃反应最佳；以乙醇为胺化反应溶剂46 ℃反应16 h最佳；S-噻吗洛尔的转晶拆分以水作溶剂，比传统不对称合成工艺安全稳定，操作简单，适合工业化生产。     

Transparent and flexible films of new segmented polyurethane nanocomposites incorporated by NH2-functionalized TiO2 nanoparticles

In this work, TiO₂ nanoparticles are surface modified by NH₂-terminated organic moieties arised from 4,4′-methylene diphenyl diisocyanate (MDI). These nanoparticles are incorporated into ether-based segmented polyurethane (SPU) matrix. MDI is utilized as monomer together with poly(tetramethylene oxide) (PTMO) comonomer for preparing the final polymer as well. The NH₂-functionalized TiO₂ nanoparticles are covalently linked to the NCO terminals of the resulting SPU macromolecules during film preparation stage. Therefore, in addition to butylene glycol, these surface modified nanoparticles with enhanced organophilicity could play the role of the second chain extender of NCO-capped SPU macromolecules through formation of urea linkages. Optical and thermal behaviors of the transparent and flexible film (SPU/TiO₂–MDI) is compared with those of unmodified TiO₂ (SPU/TiO₂) and TiO₂-unloaded SPU films. Though the particle loading is only 5 wt.%, incorporation of TiO₂ and TiO₂–MDI nanoparticles into the SPU polymer enhances significantly the light absorption in UV region at 300–400 nm. SEM images of the prepared films clearly show a considerable decrease in particle aggregation for TiO₂–MDI into SPU matrix compared to that of unmodified TiO₂. TG analyses indicate a one-step decomposition pattern with onset temperatures of about 360 and 380 °C for neat SPU and SPU/TiO₂–MDI, respectively. Moreover, DTA thermograms of both nanocomposites show obviously two exothermic phase transitions in the thermal range of 330–440 °C.     

Strain hardening behaviors and mechanisms of polyurethane under various strain rate loading

The uniaxial tension experiments are performed on thermoplastic polyurethane to investigate its mechanical behaviors and related potential mechanisms, and the loading strain rate is designing to be wide ranging from 0.0001 to 1 s⁻¹. It is found that the polyurethane presents an obvious rate-dependence, and the stress strain curves share distinct strain hardening characteristics under the investigated strain rates. Furthermore, the strain hardening ratios are sharing nearly same trends and appear to be influenced by both strain rate and the induced adiabatic heating. Besides, the ratio is also strain-dependent on previous loading history. Then, a two-dimension unit cell model is built to investigate potential equivalent mechanisms, of which the hard phase as inclusion is equivalent with crystallization zone, crosslinking sites, and so forth. The simulation results facilitate to explain the distinct strain hardening ratios, even for the matrix from the extrapolated curves under super-low strain rate loading. Finally, the analogic mechanisms of equivalent hard inclusions are proposed, which can reasonably explain the strain rate- and strain-dependence characteristics of polyurethane mechanical behaviors.     

改进HSR-FCN的服装图像识别分类算法研究

目前网络上的服装图像数量增长迅猛，对于大量服装图像实现智能分类的需求日益增加。将基于区域的全卷积网络（Region-Based Fully Convolutional Networks，R-FCN）引入到服装图像识别中，针对服装图像分类中网络训练时间长、形变服装图像识别率低的问题，提出一种新颖的改进框架HSR-FCN。新框架将R-FCN中的区域建议网络和HyperNet网络相融合，改变图片特征学习方式，使得HSR-FCN可以在更短的训练时间内达到更高的准确率。在模型中引入了空间转换网络，对输入服装图像和特征图进行了空间变换及对齐，加强了对多角度服装和形变服装的特征学习。实验结果表明，改进后的HSR-FCN模型有效地加强了对形变服装图像的学习，且在训练时间更短的情况下，比原来的网络模型R-FCN平均准确率提高了大约3个百分点，达到96.69%。