基于负相关神经网络集成的货币识别算法研究

为了提高货币识别率,提出了用负相关学习算法来提高神经网络集成的泛化能力.将紫外光照射下的纸币图片作为实验样本,将负相关学习法的集成神经网络用于分类器设计,选择6种面额纸币在不同噪声下的样本共300个作为训练样本,对单个神经网络分类器和神经网络集成分类器进行了MATLAB仿真,并对仿真所得的可靠性、识别率进行对比.实验结果表明,基于负相关学习的神经网络集成对货币识别分类有很好的效果,与应用单个神经网络的系统和独立训练个体网络的集成神经网络相比,它的识别率平均可以高出4％.

The Study of Currency Recognition Based on Negative Correlation and Neural Network Ensemble

In order to improve the recognition rate of currency, a negative correlation learning algorithm is proposed to improve the generalization ability of the neural network ensemble. The notes picture under UV-light is used as experimental samples in this paper. Ensemble neural network based on negative correlation learning algorithm is used for the classifier design. 6 kinds of denomination notes in different noise are selected under a total of 300 as the training sample. By using MATLAB to simulate the single neural network classifier and neural network ensemble classifier, the simulation of reliability and recognition rate are compared. The experiment result shows that, the neural network ensemble based on negative correlation learning is good for currency recognition classification, compared with the system using single neutral network and the integrated neural network of individual network of independent training, it has higher recognition rate of 4% in average.