基于荧光光谱技术的菜籽油氧化状态智能评价

菜籽油在加工及贮藏过程中，易受氧气、温度、光照等因素的影响，产生氧化酸败现象。为准确判断油脂氧化程度，实现不同氧化模式下菜籽油品质的快速判别，采用三维同步荧光光谱技术结合平行因子分析法及BP神经网络法建立菜籽油氧化状态的智能评价模型。以冷榨菜籽油为原料，将样品分别置于常温、Schaal烘箱、高温模式中氧化处理，期间采集菜籽油的三维同步荧光光谱数据及理化指标，当理化指标超出国标限定范围时，停止采集数据。结果表明，菜籽油中荧光物质在不同氧化模式中的演变规律呈显著差异，氧化温度对菜籽油荧光光谱有明显影响。常温氧化350 d与第1 d相比，菜籽油的特征荧光峰位置无变化，仅在激发波长Ex为620和660 nm附近荧光峰强度发生微弱变化；Schaal烘箱氧化26 d后，在激发波长Ex为620和660 nm附近荧光峰强度显著减弱，且在激发波长Ex为350～450 nm之间有新的荧光峰生成；高温氧化48 h后，Ex为620和660 nm处荧光峰消失，在Ex为400～550 nm处产生显著荧光峰，相对Schaal烘箱氧化，荧光波长发生一定程度红移，这是由于高温氧化过程中油脂氧化生成的物质稳定性较差引起的。利用平行因子分析法对三维同步荧光光谱数据进行分解获取有效的二维荧光光谱数据，当组分数为6，Δλ=60 nm时激发波长的载荷值最大，不同样品间差异最显著。选定Δλ=60 nm波段的二维荧光光谱数据用于智能评价，作为BP神经网络模型的输入值，以极性组分作为模型输出值，分别对菜籽油三种氧化模式数据建模训练。实验结果表明，三种氧化模式对应的训练集、验证集、测试集模型相关系数r均能达到0.9以上，其中常温氧化模式中验证集及测试集模型的相关系数r为1，输出值与目标值较接近，模型的预测效果较好；综合三种氧化模式数据建模，对应训练集、验证集、测试集模型的相关系数分别为0.999，0.913和0.988，均方误差均较小，说明该模型能准确判断菜籽油的不同氧化状态。因此，三维同步荧光光谱技术结合平行因子分析法、BP神经网络法建立快速检测模型能实现菜籽油不同氧化状态的判别，为菜籽油的氧化程度的评价提供新方法，同时为其他食用油的品质评价提供参考。     

基于级联卷积神经网络的前列腺磁共振图像分类

针对深度学习训练成本高，以及基于磁共振图像的前列腺癌临床诊断需要大量医学常识且极为耗时的问题，本文提出了一种基于级联卷积神经网络（Convolutional Neural Network，CNN）和磁共振图像的前列腺癌（Prostate Cancer，PCa）自动分类诊断方法，该网络以Faster-RCNN作为前网络，对前列腺区域进行提取分割，用于排除前列腺附近组织器官的干扰；以基于ResNet改进的网络结构CNN40_bottleneck作为后网络，用于对前列腺区域病变进行分类.后网络由瓶颈结构串联组成，其中使用批量标准化（Batch Normalization，BN）、全局平均池化（Global Average Pooling，GAP）进行优化.实验结果证明，本文方法对前列腺癌诊断结果较好，而且缩减了训练时间和参数量，有效降低了训练成本.     

基于多尺度学习与深度卷积神经网络的遥感图像土地利用分类

土地利用信息是国土资源管理的基础和重要依据，随着高分辨率遥感图像数据的日益增多，迫切需要快速准确的土地利用分类方法。目前应用较广的面向对象的分类方法对空间特征的利用尚不够充分，在特征选择上存在一定的局限性。为此，提出一种基于多尺度学习与深度卷积神经网络（deep convolutional neural network，DCNN）的多尺度神经网络（multi-scale neural network，MSNet）模型，基于残差网络构建了100层编码网络，通过并行输入实现输入图像的多尺度学习，利用膨胀卷积实现特征图像的多尺度学习，设计了一种端到端的分类网络。以浙江省0.5 m分辨率的光学航空遥感图像为数据源进行了实验，总体分类精度达91.97%，并将其与传统全卷积网络（fully convolutional networks，FCN）方法和基于支持向量机（support vector machine，SVM）的面向对象方法进行了对比，结果表明，本文所提方法分类精度更高，分类结果整体性更强。     

三维网络结构NiCo2S4@NiCo2O4复合电极的制备及其性能

超级电容器具有更大的功率密度、优秀的循环稳定性、极快的充放电速度、超长的循环寿命以及环境友好等突出特点,其性能与构件关系密切,其中最根本的就是组成它的电极材料。本研究主要采用传统的水热法制备出钴酸镍(NiCo₂O₄)电极材料,进而通过离子交换(二次水热)制得镍钴硫(NiCo₂S₄),最后利用化学浴沉积(CBD)法使其与钴酸镍复合,得到最终所需的三维网络结构NiCo₂S₄@NiCo₂O₄复合电极。经过表面形貌表征、循环伏安测试、恒电流充放电测试以及比电容计算分析等可以证明:三维网络结构NiCo₂S₄@NiCo₂O₄复合电极的比电容及循环稳定性等远远优于复合前单一的纯NiCo₂O₄电极材料,具有极大应用前景。     

单重工作休假M/M/c排队驱动的流体模型分析

该文在M/M/c排队驱动系统中加入工作休假策略,研究了单重工作休假多服务台排队驱动的流体模型.利用拟生灭过程和矩阵几何解法得到驱动系统稳态队长分布.构建净输入率结构,导出流体模型的稳态联合分布函数满足的的矩阵微分方程组,进而利用Laplace-Stieltjes变换(LST)方法得到稳态下缓冲器库存量的空库概率及均值表达式.最后,给出模型在多信道无线Mesh网下的应用,通过数值例子展示参数变化对系统性能指标的影响.     

Stability analysis of fractional order fuzzy cellular neural networks with leakage delay and time varying delays

In this paper we investigated the stability of fractional order fuzzy cellular neural networks with leakage delay and time varying delays. Based on Lyapunov theory and applying bounded techniques of fractional calculation, sufficient criterion are established to guarantee the stability. Hybrid feedback control is applied to derive the proposed results. Finally, numerical examples with simulation results are given to illustrate the effectiveness of the proposed method.     

Prediction of Concrete Meso-Model Stress-Strain Curves Based on GoogLeNet北大核心CSCD

非均质复合材料的宏观力学性能往往取决于细观组分的分布方式和力学性能，但是建立明确的关系表达式极其困难。为了应对这一挑战，以混凝土为研究对象，提出了一种基于深度学习的策略，能够高效、准确地通过细观模型图像信息获取应力-应变曲线。首先，使用基于卷积神经网络（convolutional neural network，CNN）的GoogLeNet模型进行图像信息识别和提取，并针对应力-应变曲线的复杂性特点，进行了数据预处理操作，并且设计了相应的多任务损失函数。数据集中的细观模型图像采用基于Monte-Carlo的随机骨料模型生成，并且使用数值模拟试验获取对应细观模型的单轴压缩应力-应变曲线。最后，通过对神经网络的训练和测试评估了所提出方法的可行性。结果表明，GoogLeNet模型训练效率和预测精度均优于AlexNet和ResNet模型，具有良好的泛化能力和鲁棒性。     

An artificial hummingbird algorithm based localization with reduced number of reference nodes for wireless sensor networks

Wireless Sensor Networks (WSN) are widely used in recent years due to the advancements in wireless and sensor technologies. Many of these applications require to know the location information of nodes. This information is useful to understand the collected data and to act on them. Existing localization algorithms make use of a few reference nodes for estimating the locations of sensor nodes. But, the positioning and utilization of reference nodes increase the cost and complexity of the network. To reduce the dependency on reference nodes, in this paper, we have developed a novel optimization based localization method using only two reference nodes for the localization of the entire network. This is achieved by reference nodes identifying a few more nodes as reference nodes by the analysis of the connectivity information. The sensor nodes then use the reference nodes to identify their locations in a distributive manner using Artificial Hummingbird Algorithm (AHA). We have observed that the localization performance of the reported algorithm at a lower reference node ratio is comparable with other algorithms at higher reference node ratios.     

Polymer networks functionalized with low-valent phosphorus cations

In this work, we show that polymer networks composed of tertiary alkyl phosphines can be cleanly functionalized with phosphino-phosphonium or triphosphenium cations. Methods for functionalizing the polymers range from halide abstraction of commercially available reagents, to ligand exchange from simple to make reported compounds, and finally, macromolecular ligand design guided by observations made at the molecular level to accommodate the formation of kinetically favored triphosphenium cation functionalized networks. The synthesis, comprehensive characterization, and comparison of the new polymers to molecular analogues is outlined. It is shown the addition of the low valent phosphorus centers to the polymer network has the effect of tuning material physical properties.     

Computational-Based Approach for Predicting Porosity of Electrospun Nanofiber Mats Using Response Surface Methodology and Artificial Neural Network Methods

Comparative studies between response surface methodology (RSM) and artificial neural network (ANN) methods to find the effects of electrospinning parameters on the porosity of nanofiber mats is described. The four important electrospinning parameters studied included solution concentration (wt.%), applied voltage (kV), spinning distance (cm) and volume flow rate (mL/h). It was found that the applied voltage and solution concentration are the two critical parameters affecting the porosity of the nanofiber mats. The two approaches were compared for their modeling and optimization capabilities with the modeling capability of RSM showing superiority over ANN, having comparatively lower values of errors. The mean relative error for the RSM and ANN models were 1.97% and 2.62% and the root mean square errors (RMSE) were 1.50 and 1.95, respectively. The superiority of the RSM-based approach is due to its high prediction accuracy and the ability to compute the combined effects of the electrospinning factors on the porosity of the nanofiber mats.