多分类器融合系统设计与应用

Design and Application of Multiple Classifier Systems

设计高性能的分类系统是模式识别研究领域追求的目标,多分类器系统MCS是实现该目标的一个有效途径。在对比MCS与单分类器系统SCS的基础上,阐述了MCS的设计与优化,并对当前的优化技术进行了分类和比较,指出了存在的问题及未来的研究方向。给出了一个用MCS实现空战目标识别的应用实例,该实例以目标的战术性能参数为分类特征,通过和规则融合多个BP网络分类器求得系统决策。实验结果表明,MCS能显著提高系统的识别率和可信度。

Design of a high performance classification system is pursued in the field of pattern recognition. Multiple classifier systems (MCS) based on combining multiple classifiers are proposed and proved to be effective in improving classification performance. After comparing with single classifier systems (SCS), a formulation of the system design of MCS as well as the system optimization is presented. Current methods for system optimization are classified and compared, then problems in them are pointed out. It also points out the research direction in MCS. A special application is given which is about the target recognition in air warfare. In the application, multiple BP classifiers are combined by the sum rule to form the MCS. Experimental results demonstrate that MCS outperform SCS in both identify ratio and reliability.