【作者】 王勇；

【导师】 潘双夏；

【作者基本信息】 浙江大学 ， 机械设计及理论， 2004， 硕士

【摘要】 纸币清分机是近年来得到快速发展的金融光机电一体化设备。论文结合纸币清分机产品的研发实践，在数字图像处理理论和当今先进的设计理论的指导下，针对现有清分机存在的不足，结合今后的发展趋势对纸币清分机的关键技术进行了较为全面深入的研究，提出了一种新的新旧识别算法，并采用最新ARM／DSP双核芯片对控制系统进行了最优化设计，从而提高了清分效果。主要涉及如下几个方面： 1．图像处理算法的设计。首先，运用数字图像处理的一些主要方法对纸币的图像特征进行了分析。在分析的基础上，提出了基于分块灰度直方图和形态区域分割的污渍特征提取新旧识别算法，由于国内类似算法具有包含新旧信息较少，分辨率不高等缺点，本算法在兼顾速度的同时，在此方面进行了较大改善，而且能够很好的识别出少量的脏点和脏线。最后，根据实际情况，选择已有的方法中较为高效的算法并进行了一些改进，实现了纸币清分机的其它功能——面额识别、版本识别和方位识别。 2．分类法及误差分析。首先介绍了系统所采用的LVQ神经网络和求相关性两种分类方法。接着，对系统新旧识别的总体误差作了分析，并在求相关性的基础上建立了衡量硬件系统图像采集效果的方法。 3．系统分析及控制策略的规划。为了提高系统的整体运行效率，首先对整个系统建立了两个简单的数学模型。接着，在这个模型的基础上，详细分析了决定系统运行速度的关键因素，并且在多处理器并行设计理论的指导下，提出了系统并行设计的几种方案。最后，阐述了纸币清分机硬件系统和软件系统具体的实现方案。由于本系统采用多处理器并行方式，采用当今先进的ARM／DSP双核芯片做主处理器，并且针对系统特点进行了优化，从而保证了系统在控制方面具有很大的优越性。更多还原

【Abstract】 Bill sorter, which is fast been developing lately, is an optical, mechanical and electrical financial facility. In this paper, on the basis of the practice of developing the bill sorter, with the digital image processing theory and the advanced design theory’s guidance, and aimed at the deficiency of the existent bill sorters and the developmental current of the bill sorter, the author makes a thorough study on the key technology, brings forward a new classification arithmetic of the new and used bills, and with the advanced chip (ARM/DSP dual kernel chip), optimizes the design of the system in the control part. Accordingly, the effect of the classification is greatly improved. The paper consists of the following parts:1. The design of the image processing arithmetic. Firstly, With the main study methods of the digital image processing, analyse the bill’s image characters. Then, on the basis, bring forward the classification arithmetic of the new and used bills based on distilling besmirchment characters by means of the histogram of image segmentation and morphogical processing of image segmentation. Because of the deficiency of including the less new and used information and the lower resolution in the native homologue, this arithmetic greatly improves in this point and primly recognizes the few dirty dots and lines, considering the velocity. At last, on the basis of the actual state, improve on the better one in the all homothetic arithmetic, consequently, realize other functions of" the bill sorter-denomination recognition, edition recognition and orientation recognition.2. The methods of the classification and the error analysis. First, introduce two classification methods adopted by this system-LVQ nerve network and curvilinear correlation. Then, analyse the composite error of the system classification and build the method, which can evaluate image collection effect of the hardware system on the basis of calculating curvilinear correlation.3. The system analysis and the layout of the control strategy. On purpose of improving the whole system productive efficiency, firstly build two simple mathematics models. Then, based on this point, analyse detailedly the key factors which determine the system running velocity. Besides, with the multiprocessor parallel design theory’s guidance, bring forward several system parallel design projects. At last, expatiate on the idiographic design projects of the hardware and software system of the bill sorter. Because of adopting the multiprocessor parallel design, adopting the current advanced chip (ARM/DSP dual kernel chip) as the main processor, and aimed at the characters, optimizing the control system, guarantee the advantage of the control system.更多还原