基于SRM自组织多区域覆盖的可拒绝近邻分类算法研究.pdfVIP

第 31 卷第 2 期 电 子 与 信 息 学 报 Vol.31No.2 2009 年 2 月 Journal of Electronics Information Technology Feb. 2009 基于 SRM 自组织多区域覆盖的可拒绝近邻分类算法研究 胡正平 贾千文 (燕山大学信息科学与工程学院 秦皇岛 066004) 摘 要：该文依据区分与划分相结合的可拒绝模式识别思路，提出了高维空间海量训练样本情况下的基于结构风险 最小化决策的自组织多区域多球覆盖可拒绝近邻分类算法。该方法利用同类样本之间相互接近的特性，通过结构风 险最小化紧几何覆盖策略，选择训练样本，通过自组织多区域多球覆盖模型构成同类样本的划分性描述，达到拒绝 识别非训练类样本的目的，最后通过k近邻相互区分性比较确定真实类别。仿真实验结果表明该文的思路是合理可 行的，在实际应用领域具有一定价值。 关键词：可拒绝模式识别；结构风险最小化原理；近邻分类 中图分类号：TP391.4 文献标识码：A 文章编号：1009-5896(2009)02-0293-04 Rejecting Nearest Neighbor Classifier Based on Structural Risk Minimization Principle Self-organization Multiple Region Covering Model Hu Zheng-ping Jia Qian-wen (School of Information Science and Engineering, Yanshan University, Qinhuangdao 066004, China) Abstract: According to the mode of rejecting pattern recognition principles, which is based on “matter description” and “matter separation” in uniform statistical pattern recognition, a rejecting nearest neighbor classifier based on structural risk minimization self-organization multiple region covering model is presented in this paper. This new model is better closer to the actual instance, rather than traditional statistical pattern recognition only using “optimal separation”as its main principle. Firstly, the optimal samples are selected from the training samples based on structural risk minimization, which is used for same class pattern matter description. Then the kNN distinguish is as a following step to identified the exact class. The simulation experimental results show that this method is valid and efficient. Key words: Rejecting pattern recognition; Structural risk minimization; Nearest neighbor classifier 1 引