基于粒子群优化的网络视频流特征选择与识别

发布时间：2018-09-05 17:20

【摘要】：随着互联网和流媒体技术的日新月异,网络视频业务流在网络中所占的比重不断增大。新兴应用业务的不断涌现使得当前的网络环境变得更加复杂,如何有效地分配有限的网络带宽资源和保障网络服务质量成为ISP(Internet Service Provider)急需解决的问题。网络视频业务流的分类技术可使ISP更好的依据不同的视频业务的QoS(Quality of Service)需求提供不同等级的服务,从而优化网络资源分配。本文主要对标清/高清/超清视频、在线直播视频、即时通信类视频、P2P类视频和HTTP视频下载七种网络视频业务进行识别研究。主要的研究工作如下:提出一种基于ReliefF和粒子群优化算法(PSO)的特征选择方法。该方法首先利用ReliefF算法滤除部分无关特征,达到快速降维的目的;然后以PSO作为搜索算法,并用ReliefF算法得到排名靠前的部分优良特征作为粒子群的初始种群,以不一致率作为评价函数在剩余特征子集中选择出最优子集。ReliefF既减小了特征空间维数,又为PSO算法提供了先验知识,从而提高了算法整体的搜索效率和分类精度。实验表明,该方法在不同的数据集上均能保持较优的性能。如何找到能反映网络视频业务本质差异的统计特征组合是网络视频业务流识别和分类的关键。通过利用本文提出的特征选择算法,可以获得一些更具有区分度的QoS统计特征组合,并且利用特征分布图验证这些特征组合的有效性。分类算法的性能除了和算法本身有关,还与所要分类的类别数目有关。分类算法的性能会随着类别数目增加而降低。因此,本文针对网络视频业务设计了一种多层SVM(Support Vector Machine)级联分类模型。每一个SVM分类器只用来识别特定类型的网络视频业务,并且每个分类器采用不同的统计特征组合。与现有方法相比,本文提出的多层SVM级联分类识别算法具有更优的分类性能。
[Abstract]:With the rapid development of Internet and streaming media technology, the proportion of network video traffic in the network is increasing. The emergence of new applications makes the current network environment more complex. How to effectively allocate limited network bandwidth resources and ensure network quality of service becomes ISP (Internet Service Provider). The classification technology of network video traffic can make ISP provide different levels of service according to different QoS (Quality of Service) requirements of video traffic, and optimize network resource allocation. Seven kinds of network video services for video download are studied. The main research work is as follows: A feature selection method based on ReliefF and particle swarm optimization (PSO) is proposed. ReliefF not only reduces the dimension of feature space, but also provides a priori knowledge for PSO algorithm, thus improving the overall search efficiency and classification accuracy of the algorithm. It is the key to identify and classify network video traffic that how to find the combination of statistical features that can reflect the essential differences of network video traffic. By using the feature selection algorithm proposed in this paper, we can obtain some more discriminatory QoS statistical feature combinations, and make use of the characteristics. The performance of the classification algorithm depends not only on the algorithm itself, but also on the number of classes to be classified. The performance of the classification algorithm will decrease with the increase of the number of classes. Each SVM classifier is only used to identify specific types of network video services, and each classifier uses different statistical features combination. Compared with the existing methods, the proposed multi-layer SVM cascade classification algorithm has better classification performance.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18;TP393.0

【相似文献】