基于BP神经网络的增益量化中的阶距自适应

本文分析了G.728语音编码算法增益固定量化的缺陷,设计了精确表示增益的自适应量化方案。采用BP神经网络对增益的量化步长进行非线性自适应量化,并提出一种新的用于训练BP神经网络的数据的获取方法。     

基于变换域注意力机制的压缩伪影去除的方法

图像有损压缩过程往往会导致图像质量退化,使图像出现压缩伪影。针对现有基于深度学习的方法缺乏对联合图像专家组(Joint Photographic Experts Group,JPEG)压缩算法先验信息的利用,提出一种基于变换域注意力机制的去伪影方法。该方法利用卷积神经网络在像素域和离散余弦变换(discrete cosine transform,DCT)域分别提取特征,再将双域学习的特征信息进行融合。利用量化表设计了DCT注意力机制,该模块根据DCT系数的损失程度给予各频率系数不同的权值,使网络自适应补偿量化引起的误差。于此基础上,在像素域引入通道注意力机制,从而更好地利用量化表的先验信息。在主要数据集上,提出的去伪影方法以固定的模型参数对多种质量因子的压缩图像进行伪影去除实验。实验结果表明,所提出的方法在各评价指标和主观视觉上取得较好的效果。     

滑模-矢量控制变频调速系统

本文介绍一种滑模-矢量控制系统。在矢量控制的基础上,采用滑模-PI控制方案,使系统获得优良的控制性能,克服了稳态时的震颠现象。系统由2片8031单片机和SPWM—GTR逆变器组成。在3kW异步电动机上进行的试验结果表明,该系统具有优良的动静态性能和较强的鲁棒性。     

负载扰动观测器在SF型矢量控制系统中的应用

文章提出广义逆矩阵来设计离散型状态观测器的方法,用该法设计的观测器,可对SF型矢量控制系统的负载力矩进行估计,再进行前馈补偿。实验证明,该法可有效改善系统的动态性能。     

一种改进的树形矢量量化编码算法

矢量量化中的一个最严重的问题是在一本码本中搜索最近码字的高计算复杂度。本文在研究树形矢量量化的基础上提出了一种改进的树形矢量量化编码算法。实验结果表明,本文提出的编码算法相对于树形矢量量化算法可大大改善峰值信噪比（ＰＳＮＲ）。     

稳定非线性动力系统周期的计算机分析

提出采用后继函数法来确定非线性动力系统的周期,并编制计算机程序对Ｖａｎ ｄｅｒ Ｐｏｌ方程实例计算周期和绘制相图,结果表明,该方法计算结果准确且实用性强,本方法不但适合于求解保守系统和具 周期解非线性动力系统的周期,也适合于耗散系统和不稳定系统的分析。     

烯烃的分子距离边数矢量(μ)及沸点的估计与预测

基于双键相对于单键键长的大小,提出了烯烃的分子距离边数（ＭＤＥ）矢量（μ矢量）;进而通过多元线性回归方法建立μ矢量与烯烃沸点之间的关系模型并进行预测,计算沸点与实验值之间的均方根误差仅为３．７８Ｋ,相关系数为０．９９９３。     

信号流图的扩展及其应用

将信号流图扩展到状态空间构成所谓向量流图,用公式可实现组合系统规范化;按工答过程绘制信号流图构成所谓过程流图,可对系统组件作深入的分析。     

Cloning, construction of prokaryotic expression vector and expression ofEscherichia coli cytosine deaminase gene

Cytosine deaminase gene ofEscherichia coli strain H-30 was cloned, and its initiation codon of ‘GTG’ was mutated to ‘ATG’ by PCR. Prokaryotic recombinant expression vector pBV220-CD was constructed. Clone with high enzyme activity were selected by detecting their specific activity of cytosine deaminase. 5-FC(5-FC, 5-fluorocytosine) could induce the lethal toxicity to cells containing active CD gene. DNA sequence analysis indicated that there were 16 altered bases and 5 of them resulted in the alteration of amino acids in predicted peptide by comparing DNA sequence of the clone H-30-CD-11 with high enzyme activity with CD gene reported in Gene Bank.     

Research advances in gene therapy for Parkinson's disease

During the long period of time when people have been seeking for an effective therapeutic method for PD, the gene therapy has shown greater and greater advantages over other methods. It can be performed mainly in two ways: ex vivo and in vivo. With the former, TH gene as well as some neurotrophic factor genes (such as GDNF, BNDF genes) can be invited to some cell lines or primary cells thus forming engineered cells and then implanting them into brain. While with the latter, viral vectors including HSV-1, Ad, AAV that can be utilized to construct recombinant viruses, or non-virus vectors can be used to delived DNA into brain directly. The present review summarizes the recent research advances in the gene therapy for PD, and it is reasonable for us to predict a notable progress in prevention and treatment for PD in the next decade.