模式识别中的支持向量机方法

Support vector machines for pattern recognition

针对模式识别问题,描述了支持向量机的基本思想,着重讨论了ν-SVM、最小二乘SVM、加权SVM和直接SVM等新的支持向量机方法,用于降低训练时间和减少计算复杂性的海量样本数据训练算法分块法、分解法,提高泛化能力的模型选择方法,以及逐一鉴别法、一一区分法、M-ary分类法、一次性求解等多类别分类方法.最后给出了污水生化处理过程运行状态监控的多类别分类实例.作为结构风险最小化准则的具体实现,支持向量机具有全局最优性和较好的泛化能力.

An overview of the basic ideas underlying Support Vector Machine (SVM) for pattern recognition was given. Methods such as v-SVM, LS-SVM, weighted SVM and direct SVM, training algorithms including chunking method and decomposition method for the sake of fast computational speed and ease of implementation, model selection approaches minimizing the generalization error, and multiclass classification methods such as one-against-the-rest method, one-against-one method, M-ary classification were concentrated. Finally, an example of multiclass classification for monitoring operation status of wastewater treatment processes was given. As a direct implementation of the structure risk minimization (SRM) inductive principle, SVM provides good performances such as global optimization and good generalization ability.