摘要
针对BP神经网络在高维数据分类中存在训练时间长的缺点,提出一种新的多神经网络分类模型,该模型采用自组织特征映射(SOFM)网络对训练样本集进行无监督聚类,通过优化竞争层神经元权值,并以此训练BP神经网络实现数据分类。最后对自由手写数字样本进行识别,仿真实验表明,这一模型具有较强的分类能力和泛化能力。
Considering the shortcoming of overlong network training time of BP neural network in multidimensional data classification, a new multiple neural network classification model is presented. Training samples are unsupervised and classified by using Self Organizing Fea- ture Map (SOFM). The weights of competition layer neurons are optimized. The BP neural network is trained with the optimized weights vector and used for data classification. This method is used for handwritten digital recognition. Simulation results indicate that the model has great capability in classification and generalization.
出处
《计算机应用与软件》
CSCD
北大核心
2007年第7期146-148,共3页
Computer Applications and Software