摘要
实际应用中的特征选择是一个满意优化问题.针对已有特征选择方法较少考虑特征获取代价和特征集维数的自动确定问题,提出一种满意特征选择方法(SFSM),将样本分类性能、特征集维数和特征提取复杂性等多种因素综合考虑.给出特征满意度和特征集满意度定义,设计出满意度函数,导出满意特征集评价准则,详细描述了特征选择算法.雷达辐射源信号特征选择与识别的实验结果显示,SFSM在计算效率和选出特征的质量方面明显优于顺序前进法、新特征选择法和多目标遗传算法.证实了SFSM的有效性和实用性.
Feature selection is essentially a satisfactory optimization problem in engineering applications. Most of the existing feature selection methods did not consider the cost of feature extraction and automatic decision of the dimension of feature subset. In this paper, a novel approach called satisfactory feature selection method （SFSM） is proposed. SFSM considers compromisingly classification performance of feature samples, the dimension of feature set and the complexity of feature extraction. Feature satisfactory rate and feature set satisfactory rate are defined. Several satisfactory rate functions are designed. Satisfactory feature set evaluation criterion is given in a mathematical way. Satisfactory feature selection algorithm is described in detail. Experimental results of radar emitter signal feature selection and recognition show that SFSM is superior to sequential forward selection using distance criterion, new feature selection method and multi-objective genetic algorithm in computing efficiency and feature qualities. Hence, the validity and applicability of the proposed method are verified.
出处
《控制理论与应用》
EI
CAS
CSCD
北大核心
2006年第1期19-24,共6页
Control Theory & Applications
基金
国家自然科学基金资助项目(60572143)
国家电子对抗重点实验室基金项目(NEWL51435QT220401)
西南交通大学博士生创新基金资助项目(2003-12)
教育部骨干教师资助计划项目(教技司[2000]65号)
关键词
优化
满意优化
特征选择
识别
optimization
satisfactory optimization
feature selection
recognition
