基于最小二乘和纠错输出编码的多类分类

Multi-classification based on least square and error-correcting output codes

多类分类是目标识别中必须面对的一个关键问题,现有分类器大都为二分器,无法满足对多类目标进行分类,为此,提出利用纠错输出编码方法对多类问题进行分解,即把多类问题转化成二类问题;同时讨论一种基于最小二乘法对二分器结果进行融合的策略。实验分别对UCI数据集和三种一维距离像数据集进行测试,结果表明与经典的多分类器相比,提出的多类分类策略有较高的分类正确率。

Multi-classification is the key issue in target recognition. The dichotomies so far is mostly designed for binary classification, which cannot meet the requirement of the multi-class target recognition. To solve this problem, the ECOC （Error Correcting Output Codes）is used to decompose a complex multi-classification problem into a set of binary classifi-cations. At the same time, a decoding strategy based on least square method is proposed to fusion the dichotomies’results. The experiments based on UCI and three kinds of different HRRPs validate that compared to the state-of-the-art dichotomies, the approach presented has better classification performance.