最小二乘支持向量机和证据理论融合的体育视频分类  

Sports video classification based on evidence theory and improved support vector machine

作　　者：曹爱春[1] 杨晓艇[2] 侯旭东 

机构地区：[1]南昌大学体育系,南昌330029 [2]江西师范大学体育学院,南昌330027 [3]上海鑫磊信息技术有限公司,上海200233

出　　处：《计算机工程与应用》2013年第23期95-99,共5页Computer Engineering and Applications

基　　金：江西省自然科学基金(No.0105100900100012)

摘　　要：针对单一特征的体育视频分类的正确率低和稳定性差等缺陷,提出一种最小二乘支持向量机(LSSVM)和证据理论相融合的体育视频分类模型(DS-LSSVM)。提取颜色、纹理、亮度、运动矢量场等4种反映体育视频类别特征,将4种单特征的LSSVM初步分类结果作为独立证据构造基本概率指派,运用DS组合规则进行决策级融合,根据分类判决门限给出最终的体育视频分类结果,最后进行仿真实验。结果表明,DS-LSSVM的体育视频分类正确率高达97.90%,相对于参比模型,DS-LSSVM具有体育视频分类正确率高、稳定性好等优势。The correct rate of sports video classification for single feature is very low and stability is poor, this paper proposes a sports video classification method combining Least Squares Support Vector Machine （LSSVM） with evidence theory （DS-LSSVM）. The color, texture, brightness, motion vector features of sports video are extracted, and then the extracted features are input into LSSVM to learn and get the preliminary classification results which are taken as evidence to establish the basic probability assignment, and DS is used to decide level fusion, the final sports video classification results are got according to the classifica- tion threshold, the simulation experiment is carried out. The simulation results show that the classification rate of the proposed algorithm reaches 97.90%, compared with the reference algorithms, the proposed algorithm has high video classification rate and good stability advantages.

关 键 词：体育视频 最小二乘支持向量机 分器设计 特征提取 证据理论 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]