【作者】 高滨；

【导师】 丁正生；

【作者基本信息】 西安科技大学 ， 应用数学， 2006， 硕士

【摘要】 支撑矢量机(SVM，Support Vector Machine)是基于统计学习理论的一种模式识别方法。使用结构风险最小化原则替代经验风险最小化原则，避免了一些长期困扰其他模式识别方法的问题，使它对于小样本学习有着较好的处理能力。利用核函数，把非线性空间的问题转换到线性空间上来解决，降低了算法的复杂度。由于具有得天独厚的优点(完备的理论基础和较好的学习性能)，使它成为当前模式识别领域研究的热点。 本文首先对SVM的理论基础—统计学习理论和相关概念进行了介绍。然后对两类SVM实现算法进行深入研究，并对他们的性能进行分析，对其优缺点进行了总结。下来对多类分类及实现算法进行了研究与分析，并对他们的算法特点进行对比。 针对大规模训练集，对SVM引入增量学习方法。这种算法通过分析SV分布的特点，采用小规模的矩阵运算来代替大规模的矩阵运算。通过对织物疵点识别的实验结果表明，该算法有效的提高了训练速度。更多还原

【Abstract】 Support vector machine is a pattern recognition algorithm based on statistical learning theory. Being substituted structural risk minimization for empirical risk minimization, support vector machine solves some problems puzzling pattern recognition field within a long period. Support vector machine has a good ability processing fewer simples. A nonlinear problem can be transformed to a linear problem with using kernel functions. The transformation reduces complexity of the algorithm. With some highlights such as perfect theories, support vector machine is a hot point in pattern recognition field nowadays.Theoretical basis for SVM-related statistical learning theory and concepts are introduced at the beginning. Some analysis of 2-category SVM algorithms performance and a summary of their advantages and disadvantages are done in chapter 2. Realization of multi-category classification algorithms be researched and analyzed in the next chapter. Some comparisons of their features are done in the chapter.For massive training sets, an incremental learning method for SVM is proposed in chapter 4. Such algorithms through analysis Support Vectors distribution characteristics use small-scale matrix operations to replace large-scale matrix operations. Experimental results of fabric blemish detection show that the algorithm effectively improves training speed.更多还原