用于图像目标识别的神经网络方法

本文提出了一种全局最优的神经网络（ＦｕｌｌＤｏｍａｉｎＯｐｔｉｍｕｍＮｅｕｒａｌＮｅｔｗｏｒｋ）模型用于目标识别。通常所设计的神经网络不能保证全局最优，使得网络不一定收敛到期望样本点上。本文的模型采用了先设计稳定点、再构造吸引域的方法，提高了网络的识别正确率及速度。针对图像识别中矢量维数大的实际，提出了一种不变性方法，使得样本维数下降而分类距离保持不变。同时又证明了网络的收敛性、收敛速度及映射保距等。计算机模拟结果表明，网络对噪声或缺损图均能正确识别。     

板坯结晶器钢水凝固的数值模拟

以实测结晶器铜板温度计算的热流量作边界条件,采用有限元方法,建立了结晶器内凝固传热方程.对凝固传热方程进行了离散化,利用ANSYS商业软件进行求解,得到凝固坯壳的应力、应变情况,从而确定连铸结晶器壁的合理锥度.     

A rapid and accurate 3-(4,5-dimethyl thiazol-2-yl)-2,5-diphenyl tetrazolium bromide colorimetric assay for quantification of bacteriocins with nisin as an example

The objective of this study is to propose a more accurate and faster MTT [3-(4,5-dimethyl thiazol-2-yl)-2,5-diphenyl tetrazolium bromide] colorimetric assay (MCA) for quantitative measurement of polypeptide bacteriocins in solutions with nisin as an example. After an initial incubation of nisin and indicator bacterium Micrococcus luteus NCIB 8166 in tubes, MTT was added for another incubation period. After that, nisin was quantified by estimating the number of viable bacteria based on measuring the amount of purple formazan produced by cleavage of yellow tetrazolium salt MTT. Then MCA was compared to a standard agar diffusion assay (ADA). The results suggested a high correlation coefficient (r ²=0.975±0.004) between optical density (OD) and the inhibitory effect of nisin on a bacterial strain Micrococcus luteus NCIB 8166 at a range of 0.125~32 IU/ml. The MCA described in this study was very quick. Quantification of nisin took only 7~8 h and the detection limit was at the level of 0.125 IU/ml when compared to 12 IU/ml and 24~28 h for ADA. The MCA provides an accurate and rapid method for quantification of nisin in solutions and is expected to be used for quantification of other antimicrobial substances.     

Amniotic membrane transplantation for porous sphere orbital implant exposure

Objective: This study is aimed at describing the clinical outcome of amniotic membrane transplantation for exposure of porous sphere implants. Methods: A retrospective review of consecutive cases of porous sphere orbital implant exposure was carried out. Eight cases were presented between May 2004 and Oct. 2006 (5 males, 3 females; mean age 44.5 years). Six had enucleation and two had evisceration. Exposure occurred in two primary and six secondary. Orbital implant diameter was 22 mm in seven cases and 20 mm in one case. Six patients are with hydroxyapatite and two with high-density porous polyethylene (Medpor) orbital implants. The mean time from implantation to exposure was 1.1 months (range 0.8~2 months). All patients required surgical intervention. Results: The time of follow-up ranged from 3.0 to 28.0 months (mean 16.5 months). Amniotic membrane grafting successfully closed the defect without re-exposure in all of these patients. The grafts were left bare with a mean time to conjunctiva of about 1 month (range 0.8~1.5 months). Conclusion: Exposed porous sphere implants were treated successfully with amniotic membrane graft in all of patients. The graft is easy to harvest. This technique is useful, dose not lead to prolonged socket inflammation and infection, and it is valuable application extensively.     

浅谈利用任务驱动法进行单片机教学与职业核心能力的培养

本文就任务驱动法在单片机教学中的应用与职业核心能力培养进行了阐述，并且以实例的形式得到了很好的验证。     

有限群极大子群的CI-截

应用有限群极大子群的CI-截性质,讨论有限群G之正规子群H的可解性,得到4个重要结论.这4个结论推广了文献[3]的相关结果.     

Preface

Different from traditional tax audit, supervision with self-audit is a combination of audit by the taxpayer himself and audit by the tax authority. This paper mainly studies a taxpayer’s optimal policy of tax evasion under supervision with self-audit and its related properties, in order to deduce some effective suggestions and theoretical bases to restrain tax evasion. Assuming that only a certain proportion of evaded tax can be discovered when the audit is executed, the authors first formulate a static model with self-audit. This model is divided into two stages. At the first stage, taxpayers declare their taxes, then the tax authority chooses audit objects, based on a known probability, and announces the result; if the taxpayer is chosen, he will enter the second stage, during which he has a chance to pay the evaded tax and the corresponding late fees and then is audited by the tax authority. The authors show the existence and uniqueness of the optimal amount of tax evasion at the first stage and the optimal proportion to self-expose at the second stage. The authors also discuss the related properties of the interior solution, and do elasticity analyses on some parameters. Besides, the authors extend the static model into the corresponding two-period model, and study the existence and uniqueness of the solution of the extended model. Finally, under the assumption that tax evasion can only be discovered with a certain probability when the audit is executed, the authors formulate another static model with self-audit and investigate its properties. This research is partially supported by the National Natural Science Foundation of China under Grant Nos. 60674082, 70221001, and 70731003.     

半正规、C-正规与有限群的超可解性

利用半正规或C-正规子群刻划有限群的结构,得到若群G的每个非循环Sylow子群的极大子群在G中或半正规或C-正规,则G超可解.     

Perceptual learning of dot pattern

Dot pattern is a type of pattern defined by specific spatial relationship among some dots. As a compensation for concrete and meaningful visual stimuli, meaningless dot pattern can be used as stimuli in learning tasks. We performed an experiment for exploring the perceptual learning process of dot pattern against random-dot background. Participants were required to learn two types of dot patterns (9-dot and 11-dot). They were assigned to two groups: participants in Group 1 learned 9-dot pattern first and 11-dot pattern later, while those in Group 2 learned 11-dot pattern first and 9-dot pattern later. The results showed that participants could acquire the spatial relationship of dot pattern through perceptual learning in relatively short learning time. In comparison with 9-dot pattern, learning time of 11-dot pattern was slightly longer and its accuracy rate lower, but there was significant positive transfer effect from 11-dot pattern learning to 9-dot learning.