论文部分内容阅读
在集成方法中，神经网络集成方法对研制集成型模式识别系统是有效的．但是，单个子分类器和集成网络的性能对集成系统的整体识别效果都有影响．因此，要进一步提高系统性能也必须改进子分类器和集成网络．文中采用竞争监督学习法，构造一个网络集成系统，用于手写数字字符识别．实验证明，该方法的确能够改进系统的收敛速度和泛化能力．
In the integrated approach, neural network integration method is effective for developing integrated pattern recognition system. However, the performance of a single sub-classifier and integrated network has an impact on the overall recognition of the integrated system. Therefore, sub-classifiers and integrated networks must also be improved to further improve system performance. In this paper, competitive supervisory learning method is used to construct a network integrated system for handwritten numeral character recognition. Experimental results show that this method can indeed improve the convergence speed and generalization ability of the system.