摘要
分析了各单一方法的分类性能，提出了神经网络与统计方法相集成的策略，由此提出MLFN-CCA-Fisher集成分类器。通过网络的自适应学习，将原样本模式经加权S型变换，映射到新的模式空间，能被线性分类，然后用相关成分分析方法提取特征，再建立Fisher判别模型。在性能测试与实际应用中，集成分类器均取得了良好的效果。
In this paper, after analyzing the classification performance of MLFN and discrimination analysis, a novel integrated classifier MLFN-CCA-Fisher is proposed. Firstly, a weighted Sigmoid transform (WS.T), through which the original patterns are mapped into a new pattern space, is obtained by self-adaptive learning of MLFN. Then the most informative classification characteristics are extracted by using the correlative component analysis. Finally, the Fisher discriminants are established. Applying the new integrated classifier,the results is satisfying not only in the performance test but also in the practical applications.
出处
《计算机工程》
CAS
CSCD
北大核心
2003年第14期47-48,185,共3页
Computer Engineering
基金
国家自然科学基金资助项目(69975017)
关键词
模式分类
集成策略
神经元网络
费希判别
相关成分
Pattern classification
Integrated strategy
Neural network
Fisher discriminant
Correlative component
