عنوان مقاله
روش مطمئن برای طبقه بندی اسکناس‌ها با استفاده از شبکه‌های عصبی مصنوعی
فهرست مطالب
مقدمه
اکتساب داده و پیش پردازش ها
استخراج ویژگی‌های تحلیل مولفه‌های اصلی
طبقه بندی
نتایج و بحث
نتیجه گیری
بخشی از مقاله
اکتساب داده و پیش پردازش ها
دو نوع حسگر برای خواندن داده استفاده می‌شوند، حسگرهای نقطه‌ای و حسگرهای خطی. داده های تولید شده توسط حسگرهای نقطه‌ای در طبقه‌بندی LVQ استفاده می‌شود، در صورتیکه حسگر خطی داده به دلیل شکل توالی آن که به عنوان یک نوع فریم دیده می‌شود؛ در HMM استفاده می‌شود.
کلمات کلیدی:
A reliable method for classification of bank notes using artificial neural networks Received: June 13, 2003 / Accepted: October 27, 2003 Abstract We present a method based on principal component analysis (PCA) for increasing the reliability of bank note recognition machines. The system is intended for classifying any kind of currency, but in this paper we examine only US dollars (six different bill types). The data was acquired through an advanced line sensor, and after preprocessing, the PCA algorithm was used to extract the main features of data and to reduce the data size. A linear vector quantization (LVQ) network was applied as the main classifier of the system. By defining a new method for validating the reliability, we evaluated the reliability of the system for 1200 test samples. The results show that the reliability is increased up to 95 % when the number of PCA components as well as the number of LVQ codebook vectors are taken properly. In order to compare the results of classification, we also applied hidden Markov models (HMMs) as an alternative classifier. Key words Bank note recognition 9 Reliability 9 PCA 9 LVQ 9 HMM 1 introduction Neural networks have been widely applied for recognition of bank notes in automatic teller machines (ATMs) in past years, and a variety of approaches have been performed to improve the classification rate and reliability of the system. L2 Due to high risk of misclassification in such systems, the reliability of recognition becomes of high importance. A. Ahmadi ([~3) 9 S. Omatu 9 T. Fujinaka Osaka Prefecture University, Department of Computer and Systems Sciences, 1-1 Gakuen-cho, Sakai 599-8531, Japan Tel. +81-72-254-9279; Fax +81-72-257-1788 e-mail: ahmadi@sig.cs.osakafu-u.ac.jp T. Kosaka Glory Ltd., Himeji, Japan This work was presented, in part, at the 8th International Symposium on Artificial Life and Robotics, Oita, Japan, January 24-26, 2003 Generally, the system must be able to classify not only different classes of different worth bills, but also the bills with different levels of fatigue, taint, and defects. Moreover, the system is expected to be robust in classification of shifted and rotated bills, and also be crucially sensitive to counterfeit notes. Concerning the complexity of the bill structure, which is used to prevent bill counterfeiting, one major step in such systems is optimizing the data acquisition and feature extraction of data. The extracted features must be discriminative enough such that they help the classifier to recognize the various kinds of input bills and also to reject counterfeit ones. Furthermore, because these machines generally work under real conditions with a wide variance in input data, a major problem is how to improve the reliability of the system to cover all kinds of real data. Thus, all proposed approaches are mainly intended to address two main problems: first, definition of a criterion for validating the reliability, and second, providing new algorithms for improving the reliability. The system proposed in this article is based on using principal component analysis (PCA) for feature extraction of data and linear vector quantization (LVQ) as the main classifier. Also, a new algorithm for evaluating the reliability of the classification is proposed, which is based on the assumption of Gaussian densities for data assigned to the codebook vectors of the LVQ, and then determination of the overlap zones. In order to have a comparative study on the results of classification, we have also applied hidden Markov models (HMMs) as an alternative classifier, which is proper for sequential input data, instead of the LVQ. Here, by using a line sensor for reading the bill data, a set of sequential frames is generated which will be used as the HMMs observation data. The experimental results obtained from different kinds of US dollar bills show that using PCA for feature extraction and LVQ as the main classifier can increase the reliability of the system significantly. The proposed method is intended for classifying different kinds of paper currency, although we have only examined US dollar bills.
