摘要
针对被污染纸币的冠字号的识别常见的几种干扰选择合适解决方法,并改进神经网络分类器,使用多层分类方法进行识别。现有的清分机和点钞机都能很好得识别八到十成新钱币的冠字号,然而,随着纸币流通不断做旧,清分机对纸币冠字号的识别明显下降,其中磨损、汗渍、折痕等是对清分机的识别影响很大的因素。本文提出多种新颖的算法,能够很好的解决被污染纸币的冠字号识别问题,并使用多层神经网络判别树架构,有效得提高了纸币识别率和容错率,最终实验结果表明,本算法能获得98.3%的识别率。
This paper chooses several solutions for the common pollution in the recognition of paper currency number, and improves the neural network classifier by using multi-layer classification recognition. The existing sorting and counting machine performs very well to recognize the eight to ten into new paper currency. However, with the circulation of paper currency and becoming old, the recognition rate of paper currency number decreases significantly, mainly because of contamination by a variety of multi-effects, such as dirt, perspiration, creases.They all affect the recognition rate of the sorter. A variety of novel algorithms are presented In this paper. The network discriminant tree branch layer is improved to advance the recognition rate and the recognition fault tolerance. The final experimental results show that this algorithm can obtain satisfactory recognition rate of 98.3%.
出处
《自动化技术与应用》
2014年第11期74-78,共5页
Techniques of Automation and Applications
关键词
斑块去除
直线去除
图形学双边缘检测
判别树
spot removal
line removal
graphics double edge detection
discriminant tree.