摘要
原始凭证的自动识别是财会信息系统的一个瓶颈。输入反例样本,分类器拒识失败是常见的错误识别类型,基于反例样本训练的神经网络分类器能够降低这类错误的发生概率。通过原始凭证的数字识别试验分析,试验数据曲线图说明采用反例样本训练的分类器能够大大地减少这种类型的错误识别,使这一类型的错误率接近于0。
Automatic recognition of Source document was the bottleneck of account information system.Artificial neural network was applied in the recognition.Negative data must be the necessary trained samples when training the classifier. The classifier could decrease some kind of recognition error.The error was that refuse recognition was not successful in inputting negative data.The experiment results showed that the classifier could deduce the error and the error rate reach the zero.
出处
《武汉理工大学学报》
EI
CAS
CSCD
北大核心
2008年第3期154-156,共3页
Journal of Wuhan University of Technology
基金
湖北省科技攻关计划资助项目(2003BDSP004)
关键词
原始凭证
反例样本
错误识别
source document
negative data
error recognition