一种基于权重融合的JPEG隐写分析方法

针对JPEG图像通用隐写检测中检测效率低、训练时间长的问题,提出一种基于集成分类器的新检测方法。算法以CC-PEV为特征对图像进行描述并作为隐写分析特征;然后,随机构造若干个特征子空间,用bootstrap方法构造图像训练子集,分别进行训练得到数个基分类器;根据基分类器的分类结果赋予基分类器不同的权重,将基分类器的结果按照其权重进行融合得到最终的结果。本文对该算法进行了测试,对它的集成性、检测准确率和训练时间进行分析。实验结果表明,相对于传统的集成方法,本文方法用自举方法构造训练集、随机方法构造子特征空间、赋予基分类器不同权重进行融合能够显著地提高算法准确率。本文方法相对于SVM和传统的集成分类方法,具有更高的检测率,对于特征维数更大的图像检测,具有更好的拓展性和一般适用性。     

基于肤色模型和改进Adaboost算法的人脸检测

为提高人脸检测的精度,提出一种基于YCbCr肤色模型和Adaboost算法的人脸检测方法。利用YCbCr颜色空间的肤色阈值模型确定待检测区域。再利用Adaboost算法准确定位出人脸位置,并对Adaboost算法的训练过程提出一种新的权值更新方法,防止权值过分增大,避免了Adaboost算法训练过程中出现的“退化”现象。     

基于变尺度混沌算法的装甲装备技术状态评估方法

针对装甲装备复杂的结构,导致其技术状态评估参数权重分配相对困难,从而影响了状态评估准确性的问题,采用变尺度混沌算法进行权重分配。首先结合该算法对技术状态综合评定参数进行了重构,并采用一般的四级技术状态指标体系对权值的分配进行分析,详细介绍利用变尺度混沌优化算法求解最优权重向量的具体过程,最后通过实例验证了该方法的合理性与有效性。     

柔性转子无试重模态动平衡测试系统

针对传统柔性转子动平衡方法需要多次添加试重、多次启动而导致平衡效率低、安全性无法保证的问题,提出了一种高速柔性转子无试重模态动平衡方法。该方法借助有限元软件计算得到转子的模态特征,结合测量得到的振动信息,计算出平衡配重的大小和相位。为了将此方法应用于工程实际,研制了一套柔性转子无试重模态动平衡软、硬件测试系统。将该系统与国外同类先进测试设备SIRIUS对比,两者之间的相位测量误差为2.376°。转速计量表明,该测试系统的转速测量范围为0～60kr/min,最大测量误差为10%。应用该系统对跨二阶双盘柔性转子实施了无试重动平衡试验,平衡后一阶振幅下降60.6%,二阶振幅下降74%。试验结果表明,所研制的无试重模态动平衡系统可有效应用于高速柔性转子的动平衡,提高平衡效率和安全性。     

Inequalities and externalities of power sector: A case of Broadlands hydropower project in Sri Lanka

The objective of the paper is to estimate environmental externalities related to a run of river project in Sri Lanka and to investigate inequity in distribution of impacts among different social groups. Diversion of the river resulted in loss of water sports (for high-income groups both local and remote), loss of historical monuments (for remote high-income groups) and recreation losses (for local poor). Removal of forest cover leads to loss of non-timber products (for local poor) and carbon storage (for remote high- and low-income groups). Loss of home garden productivity was borne by local poor groups. Benefit of the project, generation of 145 GWh annually, was a gain for the grid connected groups. The impacts were valued using various valuation methods. The base case of the cost benefit analysis resulted in NPV of US$ 11,335,730. When distributional weights are applied for different income groups, both the sign and magnitude of net benefits change. In order to be viable, the project needs diversion of at least 9% of generated electricity to the poorest households in the country. Implications for energy policy towards reducing externality and inequality impacts are also discussed.     

Studies on toxicity of multi-walled carbon nanotubes on Arabidopsis T87 suspension cells

The possible toxicities of agglomerates of multi-walled carbon nanotubes (MWCNTs) toward plant cells were evaluated. Arabidopsis T87 suspension cells were used as the model cells and decreases in values of cell dry weights, cell viabilities, cell chlorophyll contents, and superoxide dismutase (SOD) activities were seen for the cells cultured in media containing the agglomerates of MWCNTs, indicating the agglomerates of MWCNTs are toxic to the Arabidopsis T87 suspension cells. Moreover, the toxicities increased sharply as the diameters of the agglomerates of the MWCNTs became smaller. A concept based on the hypersensitive response is suggested to explain the possible toxic mechanism induced by the MWCNTs.     

导线平差取权探讨

导线平差中目前采用的取权方法有两个缺点:其一,只考虑测角和测边各自的精度比,而忽略了它们之间精度的互比;其二,未顾及测角和测边精度对导线计算的综合影响。因而导致了平差后角度改正值具有明显的系统性质。本文首先对经典取权法进行分析,指出其缺点,而后提出一组新的取权公式。最后以等边直伸导线为例导出其平差公式,并举例说明新法取权的优点。     

组合预测结果的优化

本文对组合预测法的概念及其权值的确定进行了探讨,提出了带约束的组合预测法优越性的理论依据.     

基于粗糙集理论的权重确定方法研究

针对属性权重完全未知且属性值以专家经验给出的多属性决策问题,提出了利用属性重要度计算客观权重的分配方法。根据粗糙集中的相对正域概念,探讨了如何运用属性重要度确定各属性的客观权重。决策者可以通过经验因子的选取来调整客观权重和主观权重所占的比例,通过实例说明该方法更加有效合理。     

区间数判矩阵权重求解的集成模型研究

Consistency and the weights estimation model of the interval number comparison matrix (INCM) in the analytical hierarchy process is studied under uncertainty decision-making case. Based on the weights feasible region, the local consistency definition and the local satisfactory consistency definition are given. Then, a computational model set up to test whether the INCM has the local satisfactory consistency or not. Moreover, the consistency degree based on the random crisp comparison matrix is defined as an effective index to test the consistency. Next, the upper range model, the lower range model, and the possible value model are put forward which can solve the problem that some existing approaches do not consider the consistency and its effect on the weights. According to the property of these models, a genetic algorithm is developed.