摘要
提出了一种用于图像识别的映射正交神经元网络。在一般情况下待识别的样本空间的维数相当大,为了有效地进行识别,必须把样本空间的维数降下来。目前常用的方法就是特征提取法,本文采用映射正交过程把样本空间映射成正交分类空间,并在此基础上,采用Hopfield网络进行图像分类。计算机上模拟结果表明,此网络具有对缺损和噪声图像进行正确识别的能力。
A new neural network for image recognition is proposed in this paper. In many cases of classification the number of class dimensions is much larger than that of the classes and the classes are linearly independent of each other. A ma- ppin from the class space into a new orthogonality space is used here. On the basis of mapping a Hopfield neural network used as classifer is proposed here. The results of computer simulation have revealed the high performance of the method
出处
《上海交通大学学报》
EI
CAS
CSCD
北大核心
1993年第6期14-18,共5页
Journal of Shanghai Jiaotong University
基金
国家自然科学基金
编号69072918
关键词
神经元网络
映射正交
模式识别
image recognition
neural network
mapping orthogonality
