基于SSD卷积网络的视频目标检测研究

针对传统卷积神经网络对远距离视频目标识别效果差的原因,本文提出一种改进的基于SSD卷积网络的视频目标检测模型.首先,对数据集进行剪裁,旋转等预处理,提高网络检测泛化能力,其次,采用coco数据集Mobilenet_SSD预训练模型,由于其具有轻量级网络模型特点,减少计算开销,减少内存占用量.然后,再结合voc2012数据集进行二次训练微调处理,加快训练收敛速度,使用自定义数据集能有效检测特定场景目标,能够有效识别远距离场景下视频目标物体.实验结果表明,改进的网络检测模型适用于远距离目标检测,减少计算量,降低硬件内存资源消耗,提高网络模型性能和检测精确度,具有较好的鲁棒性.     

基于深度CRF模型的图像语义分割方法

从图像中提取多种特征向量堆叠为一个高维特征向量用于图像语义分割,会导致部分特征向量的分类能力减弱或丢失。针对此问题,提出了一种结合深度卷积神经网络AlexNet和条件随机场的图像语义分割方法。利用预训练好的AlexNet模型提取图像特征,再通过条件随机场对多特征及上下文信息的有效利用来实现图像的语义分割。与利用传统经典特征的方法进行对比,实验结果表明:在利用AlexNet模型提取特征进行图像语义分割时,Conv5层为最有效的特征提取层,在Stanford background和Weizmann horse数据集下的识别准确率分别为81.0%和91.7%,均高于其他2种对比方法,说明AlexNet可以提取更有效的特征,得到更高的语义分割精度。     

Low-energy points on the sphere and the real projective plane

We present a generalization of a family of points on ${\mathbb{S}}^2$, the Diamond ensemble, containing collections of N points on ${\mathbb{S}}^2$ with very small logarithmic energy for all $N\in \mathbb{N}$. We extend this construction to the real projective plane ${\mathbb{RP}}^2$ and we obtain upper and lower bounds with explicit constants for the Green and logarithmic energy on this last space.     

Accuracy and CPU time of JE-TLM,PLRC-TLM and CRC-TLM methods for plasma medium

The current density (JE), the piecewise linear recursive convolution (PLRC) and the constant convolution recursive (CRC) techniques are developed and integrated into the transmission line matrix (TLM) algorithm. A comparison of these three schemes according to the criteria of accuracy and CPU time is presented. Numerical experiments show that JE provides the most accurate solution and requires the least CPU time; PLRC is nearly as accurate but consumes more CPU time than CRC.     

How good is CuAAC “click” chemistry for polymer coupling reactions?

¹H NMR and SEC analyses are used to investigate the overall efficiency of Copper Catalyzed Azide Alkyne Cycloaddition (CuAAC) “click” coupling reactions between alkyne‐ and azide‐terminated polymers using polystyrene as a model. Quantitative convolution modeling of the entire molecular weight distribution is applied to characterize the outcomes of the functional polymer synthesis reactions (i.e., by atom transfer radical polymerization), as well as the CuAAC coupling reaction. Incomplete functionality of the azide‐terminated polystyrene (∼92%) proves to be the largest factor compromising the efficacy of the CuAAC coupling reaction and is attributed primarily to the loss of terminal bromide functionality during its synthesis. The efficiency of the S_N2 reaction converting bromide to azide was found to be about 99%. After taking into account the influence of non‐functional polymer, we find that, under the reaction conditions used, the efficiency of the CuAAC coupling reaction determined from both techniques is about 94%. These inefficiencies compromise the fidelity and potential utility of CuAAC coupling reactions for the synthesis of hierarchically structured polymers. While CuAAC efficiency is expected to depend on the specific reaction conditions used, the framework described for determining reaction efficiency does provide a means for ultimately optimizing the reaction conditions for CuAAC coupling reactions. © 2017 Wiley Periodicals, Inc. J. Polym. Sci., Part A: Polym. Chem. 2018 , 56, 75–84     

Deflagration-to-detonation transition process for spherical aluminum dust/epoxypropane mist/air mixtures in a large-scale experimental tube

The deflagration-to-detonation transitions (DDTs) for clouds of spherical aluminum dust (SAD) mixed with air or epoxypropane mist (EPM) and air were investigated in a 29.6-m-long experimental tube of 199 mm in diameter. The clouds formed through the injection of SAD and SAD/liquid epoxypropane samples into the experimental tube. Explosions of the SAD/air mixture were initiated using a 7-m-long EPM/air cloud explosion ignited by a 40-J electric spark. Explosions in SAD/EPM/air clouds were initiated using a 1...     

Finite tchebycheff transformations of generalized functions

By using Hilbert space techniques, the finite Tchebycheff transformations of the first and the second kind are defined on appropriate spaces of generalized functions. The inversion theorems are established and some operation transform formulas are obtained. These results are applied in solving certain distributional differential equations.     

面向同源蛋白质探测的一种新型混合深度学习模型

根据蛋白质氨基酸链探测其同源蛋白质,进而预测蛋白质的功能,是生物信息学研究领域的一个重要挑战,也是众多生物医学研究领域的基础研究内容,有着重要的科研价值和广泛的应用需求。其研究难点在于:(1)如何学习对同源蛋白质预测有效、有用的蛋白质特征信息;(2)如何更好地运用蛋白质特征信息,实现同源蛋白质的探测与识别。为了解决同源蛋白质探测与识别研究中的关键难点,本文提出一种基于混合深度学习架构的同源蛋白质探测与识别模型(HDLM-PHP)。通过采用统一的"管道式"深度学习架构,将蛋白质特征学习和探测识别统一为一个整体,提高同源蛋白质探测与识别的效能。采用多组并行的深度卷积神经网络,学习蛋白质的各种属性信息,以期获得丰富的待检测蛋白质和靶蛋白质的高级相关性特征,并通过全连接方式使用多层RBM结构融合和精炼这些相关性特征为全局相关性特征。通过统一的深度网络连接方式,以探测和识别任务为导向,学习到对于同源蛋白质预测最有效、最全面的蛋白质特征信息。在标准数据集SCOPe上,对所提模型进行性能与效率评测,结果表明:本文提出的模型能有效地学习到符合任务导向的蛋白质特征数据,提升同源蛋白质探测与识别的准确度和召回率,优于现有的模型和算法。     

Characteristic behaviour of plasma fluctuations inside a plasma bubble in presence of a magnetic field due to the formation of a potential well

Experiments were carried out to observe the effect of a magnetic field and grid biasing voltage in presence of a plasma bubble in a magnetized, filamentary discharge plasma system. A spherical mesh grid of 80% optical transparency was negatively biased and introduced into the plasma for creating a plasma bubble. Diagnostics via an electrical Langmuir probe and a hot emissive probe were extensively used for scanning the plasma bubble. Plasma floating potential fluctuations were measured at three different positions of the plasma bubble. The instability in the pattern showed the dynamic transition from periodic to chaotic for increasing magnetic fields. Time scale analysis using continuous wavelet transform was carried out to identify the presence of non‐linearity from the contour plots. The mechanisms of the low‐frequency instabilities along with the transition to chaos could be qualitatively explained. Non‐linear techniques such as fast Fourier transform, phase space plot, and recurrence plot were used to explore the dynamics of the system appearing during plasma fluctuations. In order to demonstrate the observed chaotic phenomena in this study, characteristics of chaos such as the Lyapunov exponent were obtained from experimental time series data. The experimentally observed potential structure is confirmed with numerical analysis based on fluid hydrodynamics.