摘要
本文提出一种推广的三层神经网络方法,可用于线性相关样本三维目标的模式识别。该方法采用矩阵理论中的奇异值分解技术来求解样本集的综合鉴别函数(SDF),并以此FDF做为第一级神经网络的互连权重。现有的SDF方法要求训练样本集线性非相关,而我们提出的方法则取消了这一限制条件,使该网络的应用范围更为广泛。
In this paper, an extended three-layer neural network is proposed for classifications with linearly correlative training set. The singular value decomposition technique in matrix theory is adopted to obtain the synthesis discrimination functions (SDFs) of training set, and the SDFs are used as the interconnection weights of the first neural network. The limitation of noncorrelative training set in the existing SDF method is eliminated by the proposed technique.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
1994年第3期203-207,共5页
Pattern Recognition and Artificial Intelligence
基金
国家自然科学基金
教委博士点基金
关键词
模式识别
神经网络
综合鉴别函数
Neural Network Interconnection Weights, Synthesize Discrimination Function (SDF), Singular Valuie Decomposition,Pattern Recognition.
