基于“快速投票”算法的HMM/SVM混合识别模型及应用  

An HMM/SVM Mixed Recognition Model Based on “Fast Voting ” Algorithm and Application

作　　者：罗泽举[1] 朱思铭[2] 

机构地区：[1]重庆工商大学计算机科学与信息工程学院,重庆400067 [2]中山大学数学与计算科学学院,广州510275

出　　处：《计算机科学》2007年第5期215-217,227,共4页Computer Science

基　　金：国家自然科学基金资助项目(No.10371135)

摘　　要：提出一种基于隐马尔可夫模型(HMM)和支持向量机(SVM)的双层过滤识别系统。根据隐马尔可夫模型训练中不同结构的序列其L值分布范围不同的特点,对传统多类“投票模型”进行改进,提出一种“快速投票”算法。先用HMM对人类内含子和外显子进行识别,同时,对于L值区域有重叠造成识别率较低的部分,再用支持向量机进行第二次识别过滤。这一模型克服了传统用单一HMM识别方法的不足,实现了HMM和SVM的优势互补。实验表明,用HMM/SVM进行两类识别,其平均识别率达到了90%,进行多类识别,平均识别率达到了91.5%。Propose a kind of HMM/SVM double layer filter recognition system. According to the characteristic that the L value is different while the sequence structure is different in the HMM training, improve the traditional ＂voting model＂,put forward a ＂Fast Voting＂ algorithm. First, use HMM models to recognize human intron and exon,meanwhile, for the part that L value range has part overlapped,use SVM for the second recognition. This model overcomes the deficiency of the traditional single HMM recognition method, realizes the mutual supplement with each other＇s advantages of HMM and SVM. Experiment indicates that carrying on recognition of two kinds of questions with HMM/SVM, the average rate of accuracy is up to 90%, carrying on multiple classifying, the average rate of accuracy is up to 91. 5%.

关 键 词：HMM/SVM模型 “快速投票”方法 内含子和启动子识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]