嵌入式纸币识别系统

发布时间：2018-05-02 19:13

本文选题：纸币面值识别 + 序列号识别　；参考：《北京邮电大学》2012年硕士论文

【摘要】：随着计算机技术的高速发展,自动化、智能化系统在各行业内的应用越来越广泛。智能纸币识别系统可以自动完成对纸币进行鉴伪、按质量分类、面值识别和序列号识别等工作,成为银行业的一大助力。然而,目前功能强大的纸币识别系统往往是大型设备,体积大,功耗大,造价也高。因此,可移动、低功耗和高性价比的嵌入式纸币识别系统将受到中小企业的欢迎。本文以此为目标,研究和设计嵌入式纸币识别系统。本文对现有纸币识别算法进行了大量的调整和改进：提出了一种全局阈值和局部分布特点相结合的二值化方法,并采用边缘图像二值化和分段二值化等策略对原算法中多处二值化算法进行了有针对性的调整,以适应纸币图像的特点；提出一种新的基于模板匹配的序列号定位算法,以解决原算法效率低下和准确率不足的问题；另外,本文对纸币边界定位、图像噪声处理和序列号字符分割等多个算法进行了调整。经过以上改进工作,纸币定位准确率由93%提高到100%,序列号字符分割准确率也有了很大提高。本文选择了性价比较高的DM642作为嵌入式系统的处理器,并完成了纸币识别算法向嵌入式平台的移植工作,用C语言重写了STL中的Vector类和部分OpenCV函数。为了提高处理速度,本文亦对系统进行了多层次的优化工作,包括C代码优化与DSP平台相关优化。经过优化,本文使面值识别和序列号识别的时间都从1000ms以上降到20ms以下,达到实时处理的标准。
[Abstract]:With the rapid development of computer technology and automation, intelligent system is more and more widely used in various industries. Intelligent banknote recognition system can automatically complete the identification of banknotes, according to the quality of classification, face value recognition and serial number identification, become a great help of the banking industry. However, the current powerful banknote recognition system is often a large equipment, large volume, high power consumption and high cost. Therefore, mobile, low-power and high-cost-effective embedded banknote recognition system will be welcomed by small and medium-sized enterprises. This paper studies and designs an embedded banknote recognition system. In this paper, a large number of adjustments and improvements have been made to the existing banknote recognition algorithms: a binarization method combining the global threshold and the local distribution characteristics is proposed. In order to adapt to the characteristics of banknote image, a new sequence number location algorithm based on template matching is proposed, which adopts the strategies of edge image binarization and segmented binarization to adjust the binarization algorithm of the original algorithm to suit the characteristics of the paper currency image. In order to solve the problems of low efficiency and low accuracy of the original algorithm, in addition, the paper adjusts several algorithms, such as border location of paper currency, image noise processing and serial number character segmentation. After the above improvement, the accuracy rate of banknote positioning is improved from 93% to 100%, and the accuracy of serial number character segmentation is also greatly improved. In this paper, DM642 with high performance-to-price ratio is chosen as the processor of embedded system, and the paper currency recognition algorithm is transplanted to embedded platform. The Vector class and some OpenCV functions in STL are rewritten with C language. In order to improve the processing speed, this paper also carries on the multi-level optimization work to the system, including the C code optimization and the DSP platform correlation optimization. After optimization, the time of face value recognition and serial number recognition is reduced from above 1000ms to below 20ms, which is up to the standard of real time processing.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP391.41;TP368.1

【引证文献】