Development of a fish‐based index of biotic integrity (FIBI) for monitoring riverine ecosystems in the Lake Victoria drainage Basin,Kenya

The long term monitoring of aquatic ecosystems based on chemical analysis is expensive resulting in a lack of time series data for management purposes thus necessitating the need for new cost effective methodologies. This study developed a fish‐based index of biotic integrity (FIBI) for monitoring environmental conditions in riverine ecosystems within the Lake Victoria Basin, Kenya. The fish fauna in the basin is poorly studied, less diverse and ecologically specialized than the temperate fauna that provided the origins of the Index of Biotic Integrity. Fish samples were collected by electrofishing from 24 sites during baseflow periods in February, March and July 2004. Validation data were collected from 7 sites with varying levels of degradation. Fish samples were identified, counted and grouped into different trophic groups, relative tolerance to pollution, habitat guild and whether exotic or native to riverine environment. Thirty‐three candidate metrics were evaluated for responsiveness to habitat quality and twelve were selected for inclusion in the final index. The index classified 6 of the 7 validation sites according to their levels of degradation. As a bioassessment tool, the index was useful in laying the basis for long‐term monitoring of rivers in the Lake Victoria drainage basin. Copyright © 2010 John Wiley & Sons, Ltd.     

一种基于信息度量的流特征遴选算法

提出一种基于信息度量的流特征选择算法。该算法可分为粗粒度选择和细粒度选择2个选择步骤。粗粒度的选择通过计算特征集合中各个特征与不同业务类别的互信息,选择在流分类中最具代表性的特征。对于选取的这些特征进行细粒度的选择处理,通过计算已选特征集合中特征之间的一致性,排除多余的特征。实验结果表明,该算法遴选出的特征在用于数据流分类时,准确率和召回率都较同类算法高,且时间复杂度较低。     

基于黎曼流型度量的人工鱼群算法视觉跟踪

针对经典的基于协方差算子的跟踪方法不能适应目标的遮挡及其全局搜索造成的过多计算消耗问题,提出了一种在黎曼流型度量上的人工鱼群算法的视觉跟踪方法。该方法将融合了目标的位置、颜色、梯度等特征区域的协方差算子作为目标的表观模型,以提高它对姿态变化以及亮度变化的适应性。利用人工鱼群算法搜寻目标与候选目标之间最优的匹配,其并行运算机制提高了跟踪算法的效率,其全局搜索的能力则提高了算法对遮挡问题的鲁棒性。实验结果表明,该算法在复杂背景情况下具有目标跟踪的鲁棒性。     

采用加权协方差矩阵描述子的OCR识别方法

传统光学字符识别（Optical Character Recognition,OCR）方法一般只提取图像亮度特征,在图像退化较严重时识别准确率不高。针对这一问题,提出一种新的扫描字符特征提取方法。除各通道亮度外,还提取像素位置、亮度的一阶导、二阶导等特征构成特征图像,并根据各个特征对图像的贡献程度进行加权处理。计算以当前像素为中心的局部区域特征图像块的协方差矩阵作为当前像素的描述子,然后在黎曼空间对字符实施分类。实验结果表明,采用典型的结构化分类器时,该特征提取方法对字符识别的准确率高于传统方法,表现出较强的鲁棒性。     

基于NLA的Polymorphic蠕虫特征自动提取算法研究

随着Polymorphic蠕虫变形技术的不断发展,如何快速有效地提取其特征是入侵检测中特征提取领域的一个重要研究方向。采用基于模式的特征提取算法NLA(Normalized Local Alignment),通过对多个可疑Polymorphic蠕虫流量进行序列比对,自动提取高相似度公共子序列,以向量的形式构造蠕虫特征。实验结果表明该算法在误报率和漏报率方面均优于传统算法。     

Power-conserving routing of ad hoc mobile wireless networks based on entropy-constrained algorithms

This paper presents an approach to power-conserving routing of ad hoc mobile wireless networks. This approach relies on entropy-constrained routing algorithms, which were developed by utilizing the information-theoretic concept of the entropy to gradually reduce the uncertainty associated with route discovery through a deterministic annealing process. Entropy-constrained routing algorithms were tested using a single performance metric related to the distance between the nodes and to the power consumption associated with packet transmission. This paper also expands the versatility of entropy-constrained routing algorithms by making them capable of discovering routes based on multiple performance metrics. In this study, the second performance metric employed for route discovery relied on the power availability in the nodes of the network. The proposed routing approach was evaluated in terms of the power consumption associated with the routing of packets over an ad hoc mobile network in a variety of operating conditions.     

基于自适应距离度量的最小距离分类器集成

提出了一种基于自适应距离度量的最小距离分类器集成方法，给出了个体分类器的生成方法。首先用Bootstrap技术对训练样本集进行可重复采样，生成若干个子样本集，应用生成的子样本集建立自适应距离度量模型，根据建立的模型对子样本集进行训练，生成个体分类器。在集成中，将结果用相对多数投票法集成最终的结论。采用UCI标准数据集实验，将该方法与已有方法进行了性能比较，结果表明基于自适应距离度量的最小距离分类器集成是最有效的。     

基于度量的决策树

决策树是一种采用分治策略的聚类分析方法，构建决策树的关键是选择合适的属性。传统的决策树通常从最大化信息熵的角度来构造，不能对属性的分类能力进行足够好的区分。对传统的决策树生成算法的不足，本文提出了一种基于马氏距离的决策树生成算法。算法使用马氏距离来区分不同特征属性子集的分类能力。试验结果表明，基于度量的决策树的性能优于传统的决策树。     

网络主机脆弱性分析

本文提出了一种基于Agent的网络主机脆弱性分析框架，通过分布在各主机中的Agent对网络主机脆弱性度量进行收集，然后分析在不同网络攻击阶段的脆弱性度量值变化状况。通过该脆弱性评估框架，不仅分析了网络主 机的脆弱性状况、模拟了网络攻击过程，而且为进一步的网络安全防范奠定了基础。     

一种基于可用带宽的QoS多播路由算法

设计一个加权最大可用带宽多播树求解模型,将可用带宽作为主度量并同时兼顾时延、时延抖动和包丢失率3个QoS约束构建多播树.各QoS度量的重要性可根据实际的服务需求灵活调整,算法具有多项式复杂度,并通过仿真证明其有效性.