一种改进的AdaBoost算法——M-Asy AdaBoost  

A Revised AdaBoost Algorithm—M-Asy AdaBoost

作　　者：张彦峰[1] 何佩琨[1] 

机构地区：[1]北京理工大学信息与电子学院,北京100081

出　　处：《北京理工大学学报》2011年第1期64-68,73,共6页Transactions of Beijing Institute of Technology

摘　　要：提出一种Asymmetric AdaBoost改进算法——M-Asy AdaBoost.M-Asy AdaBoost算法通过新的样本权重分配方式可以确保训练过程不失败;分类器权重采用对正样本的分类错误率形成优化权重,突出对正样本的识别能力,提高检测概率;并且通过对加入分类器集的分类器的限制,使检测概率单调增加.该算法在较低虚警概率下,达到高检测概率.计算机仿真结果验证了算法的正确性.This paper presents a revised type of Asymmetric AdaBoost algorithm--M AdaBoost. It can ensure the success of training process by sample weight distribution. Asy The weight of classifier is optimized by adopting error rates of positive sample. The capability of recognizing positive sample is enhanced. The detection probability is improved and increased monotonously by restricting the classifier adding in the ensemble. The proposed algorithm can attain high detection rate with low false alarm ratio. In the end, it is proved by computer simulation that the M-Asy AdaBoost is effective.

关 键 词：M-Asy ADABOOST 分类器 分类器集 ASYMMETRIC ADABOOST 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]