用于七段码识别的BP网络中虚拟样本的生成

针对BP网络用于工业仪表识别中的泛化能力较差问题,根据七段码数显仪表的特点,采用模拟七段码码段重叠的方法,构造虚拟样本,并研究了包含虚拟样本的新训练集和不包含虚拟样本的原始训练集下BP网络的识别效果和泛化能力,发现了新训练样本集训练得到的BP网络对仪表上的七段码识别率与原始训练集训练得到的BP网络相比,识别效果较好,且泛化能力提高了。实验结果表明了该虚拟样本的构造方法的有效性和合理性。

Generation of virtual sample used for the seven-segment code identification in BP network

Aiming at the poor generalization ability of the BP network used in industrial instrumentation identification,a new method that constructs virtual samples by using seven-segment code yards segment method is proposed.The BP network trained with the new training set which contains virtual samples is studied,and it discovers that it has better generalization ability than the BP network trained with the original training set,which does not contain the virtual samples.The experimental results show that the constructing method of the virtual samples is effective and reasonable.