【作者】 唐晓东；

【导师】 徐东平；

【作者基本信息】 武汉理工大学 ， 计算机应用技术， 2007， 硕士

【摘要】 随着信息技术的发展，图像处理与识别技术己应用于例如交通管理、银行支票识别、医学图像中的癌细胞识别、遥感技术等许多领域，成为21世纪具有时代特征的重要技术之一。本文从图像处理与识别技术的理论知识出发，详细阐述了“基于BP神经网络的集装箱箱号识别系统”的研究与开发内容以及系统处理流程：图像预处理、箱号字符区域定位及字符分割、特征提取和字符识别。在整个设计过程中由于现有的算法不能很好的适应工程实践的具体要求，因此在设计的各个主要环节都不同程度的提出了改进的算法，主要工作如下：1．在集装箱图像预处理方面，提出了采用中值滤波和去杂点法相结合的方法去除噪声。2．在箱号定位方面，针对集装箱箱号字符特点提出了采用频率法并结合以箱号位置大小为先验知识的定位方法。3．在箱号识别方面，利用BP神经网络算法设计字母网络和数字网络分别对集装箱箱号中的字母和数字进行识别，避免相近字符混淆。最后，应用所提出的方法，在Microsoft Visual C++6.0开发平台下针对所采集到的图像进行了识别实验，实验结果表明，灵活的运用图像处理和识别技术可有效的识别集装箱箱号，识别率可达96％以上，具有实际应用的前景。更多还原

【Abstract】 With the development of Information Technology, the image processing and recognizing technology have already applied to a lot of fields , just as traffic control, bank check recognition, cancer cell distinguishing , remote sensing technique, and so on and becomes one of the important technologies having features of the times on the new 21 centuries .Staring from the theory of image processing and recognizing technology, this thesis expatiate the content of research and development about "container’s box number recognize system based on BP neural network" detailedly and system handling process: image preprocess, box number area allocation, character segmentation, feature extraction and recognition. Because present algorithms cannot perform the project practice’s idiographic request perfectly, therefore bring forward the improved algorithm on varying degrees in each mostly tache about designing. Main work as follows:1. In container’s image preprocessing, bring forward use the methods of median sieve combine with remove scatter noise method to cancel the noise.2. In box’s number location, according to container’s box number character characteristic use the location methods of frequency combine with the prior knowledge of box’s number position.3. In box’s number recognition, design respectively letter network and digital network to recognize the box’s number of letters and digits by BP nerve-network to avoid similar character mix up.Finally apply what be suggested that method, we take an experiment on acquired images with Microsoft Visual C++6.0 platform. The experiment result indicates that using image processing and pattern recognition technique flexibility can recognize container’s box number, the recognition rate can be more than 96 percent. It has prospect of practical application.更多还原