基于RBF网络的增量学习

提出了一种基于径向网络的增量学习算法 ,在对新样本进行学习时 ,通过恢复受干扰样本的方式来尽可能保持已有的知识不变 .模拟实验证明了本算法的正确性和有效性 .     

上游无情报的水文预报技术初探

以珠江水系上游左江的龙州水文站为例，对水文站网或是通讯设施跟不上的上游无情报提供的流域，进行水文预报方法的探讨，寻求增长中下游洪水预报预见期。     

一种自适应PID控制系统

本文根据Ｗｉｄｒｏｗ－ＨｏｆｆＬＭＳ算法，对数字ＰＩＤ增量型控制算法进行分析，导出一种具有自学功能的自适应ＰＩＤ控制系统，使其能自动调整最佳控制参数．     

渣浆泵过流件用高铬铸铁材料特性分析与选择

基于定量金相分析和Thermo-Calc软件的热力学平衡相计算,研究了渣浆泵过流件用高铬铸铁中碳化物数量和基体铬含量与铸铁中碳、铬含量之间的定量关系,在此基础上阐明了各类高铬铸铁的主要特性及适用范围。结果表明,在不同工矿条件下渣浆泵过流件最佳使用效果可通过高铬铸铁中碳和铬含量的调整来实现,高铬铸铁中较高的碳含量有利于改善其耐磨性能,而高铬低碳配置有助于获得具有良好耐腐蚀性能的不锈钢基体。Thermo-Calc软件预测结果与定量金相分析结果基本上相符,可作为各类高铬铸铁材料特性分析及选用的工具。     

一种有效的运动目标辨别方法

提出了一种利用图像序列中两帧相邻图像从分割所得二值图像的各个候选目标块中确定真实目标的有效算法．背景配准后，可获得背景噪声分布的直方图，同时获得具有对应关系的候选目标块间的噪声分布直方图，通过比较它们与背景噪声分布直方图之间的差异性，即可准确地辨别出真实目标．通过对实际图像序列进行实验，证实了本方法的有效性．     

Glycine binding primes NMDA receptor internalization

NMDA (N-methyl-d-aspartate) receptors (NMDARs) are a principal subtype of excitatory ligand-gated ion channel with prominent roles in physiological and disease processes in the central nervous system. Recognition that glycine potentiates NMDAR-mediated currents as well as being a requisite co-agonist of the NMDAR subtype of 'glutamate' receptor profoundly changed our understanding of chemical synaptic communication in the central nervous system. The binding of both glycine and glutamate is necessary to cause opening of the NMDAR conductance pore. Although binding of either agonist alone is insufficient to cause current flow through the channel, we report here that stimulation of the glycine site initiates signalling through the NMDAR complex, priming the receptors for clathrin-dependent endocytosis. Glycine binding alone does not cause the receptor to be endocytosed; this requires both glycine and glutamate site activation of NMDARs. The priming effect of glycine is mimicked by the NMDAR glycine site agonist d-serine, and is blocked by competitive glycine site antagonists. Synaptic as well as extrasynaptic NMDARs are primed for internalization by glycine site stimulation. Our results demonstrate transmembrane signal transduction through activating the glycine site of NMDARs, and elucidate a model for modulating cell-cell communication in the central nervous system.     

Modification of the rib operon derived from Bacillus subtilis and its expression in Escherichia coli

A riboflavin operon （rib operon） derived from Bacillus subtilis 368 was modified on structure and the resulting operons were expressed in various strains of Escherichia coli. The results showed that the optimization of the rib operon and the host strain used for expression are two main factors affecting the riboflavin production. Replacing the promoterl and rfn box of the rib operon with a strong constructive promoter spol drastically increased the expression of the rib genes. When E. coli JM109 was used as the host strain, the highest riboflavin production reached 95.3μg/mL （about eight times higher than that of the unmodified r/b operon）. In addition, when tetracycline （20 μg/mL） was used as the selective pressure, compared with the ampicillin resistant transformants, a higher riboflavin yield was obtained in tetracycline resistant host strain.     

陕西省兰科植物种类及其分布研究

对陕西省兰科植物种类及分布进行了调查研究.调查地点涉及兰科植物分布相对集中的30个县以及植物物种多样性相对丰富的15个自然保护区,野外调查在不同季节进行;分别对陕西省兰科植物进行了资源调查、标本采集和图片拍摄.对采集的标本通过查阅相关文献和与相关植物标本馆(室)兰科植物标本进行对比研究,共发现陕西省野生兰科植物41属82种,属、种数比以往的记录有大幅度增加.在此基础上,对陕西省兰科植物资源的合理开发、保护和管理提出了建议.     

视觉感知结合学习的自然图像轮廓检测

本文结合非经典感受野的视觉特性与机器学习的方法,提出了一种自然图像轮廓检测模型.当非经典感受野中的刺激与感受野中心刺激形成一种精确的空间结构时,将对中心产生一种增强效应;另一方面非经典感受野中抑制作用会降低同质成分的响应,我们将这两个机制分别用于增强光滑的轮廓和减少背景中与结构无关的干扰成分.利用逻辑回归概率模型将感受野中的信息与来自非经典感受野中的信息进行有效融合,并根据图像的手工标注数据库,通过学习方法获得一组最优的模型参数.自然图像的实验结果表明该轮廓检测方法能极大地抑制来自纹理的局部边缘,减少虚假轮廓,同时能增强具有一致空间结构的成分,避免轮廓缺失.最后利用Berkeley图像数据库定量地评价了我们方法的性能,并与相关方法进行了比较.该模型不仅为复杂场景中的轮廓检测提供了一个可行的策略,并有助于对生理视觉机制的理解.