Abstract: Euclidean distance is used as a vector similarity measure function in LVQ and GLVQ, which neglects the differences of feature data range and affects classification accuracy of them. Aimed at this problem, a kind of vector similarity measure function with feature data range taken into account is proposed, and a new algorithm named as GLVQ-FR based on this measure function and GLVQ is put forward. Using 8 data sets of the UCI machine learning repository, the comparative experiments of the GLVQ-FR with the LVQ2.1, GLVQ, GRLVQ and GMLVQ algorithms are carried out, results show that the classification accuracy and computation speed of GLVQ-FR algorithm are higher than the others. The algorithm usability and high performance in real production environment is verified through the video vehicle classification data set.

Key words: vehicle classification, Learning Vector Quantization（LVQ）, similarity metric, pattern recognition

摘要： 学习向量量化（LVQ）和泛化学习向量量化（GLVQ）算法都是采用欧氏距离作为相似性度量函数, 忽视了向量各维属性的数据取值范围，从而不能区分各维属性在分类中的不同作用。针对该问题，使用一种面向特征取值范围的向量相似性度量函数，对GLVQ进行改进，提出了GLVQ-FR算法。使用视频车型分类数据进行改进型GLVQ和LVQ2.1、GLVQ、GRLVQ、GMLVQ等算法的对比实验，结果表明：GLVQ-FR算法在车型分类中具有较高的分类准确性、运算速度和真实生产环境中的可用性。

关键词: 车型分类, 学习向量量化, 相似性度量, 模式识别