Current Issue Cover
欧元硬币年份检测与识别方法

郭雪峰①②,陈红磊①②,张东波①②(湘潭大学 信息工程学院)

摘 要
目的:硬币的年份是判别硬币外观质量的一个重要信息,为了对流通中的欧元硬币进行准确的清分,有必要对欧元硬币上的发行年份进行检测与识别。但由于欧元硬币年份数字的位姿不确定性,尺寸的非归一化,其它文字符号的干扰,数字排列方式的多样性使得利用计算机视觉算法实现欧元硬币年份的自动检测、识别与判读存在较大的困难。方法:针对欧元硬币年份检测与识别的特殊性,提出基于Faster-RCNN模型的数字检测方法,以及基于聚类算法和先验规则的年份排序算法。通过训练数据增量化处理,例如旋转、缩放等方式极大的扩充训练样本的规模;重新训练的Faster-RCNN网络模型能够适应硬币中数字的各种位姿和尺寸变化,进而利用K-means聚类算法,可以将获得的数字候选框聚成4类,选取每类中置信度最大的候选框,最后根据预先确定的不同国别硬币的年份排列方式,通过适当的排序算法即可得到正确的年份信息。结果:在自建的实验平台上对12个欧盟国家的5种较大币值的硬币进行采样获得4429张图片,按1:1比例划分训练样本和测试样本,实验表明,本文方法的年份检测识别准确率达到89.62%,计算耗时约215ms,基本满足准确性和实时性要求。结论:算法具备实时、鲁棒、高精度的良好性能,具有较高的实际应用价值。
关键词
Year detection and recognition of euro coin

GUO Xuefeng①②,CHEN Honglei①②,ZHANG Dongbo ①②(The College of Information Engineering,Xiangtan University)

Abstract
Objective In the circulation process, the appearance quality of coins decreases due to wear, so it is necessary to recycle the badly worn appearance. It is a common practice to judge whether the coins should be recycled by evaluating the appearance quality. The year of coins is an important information to judge the appearance quality of coins. In order to accurately identify the euro coins in circulation, it is necessary to detect and identify the year of issue on the euro coins. However, due to the uncertainty of the position and posture of the Euro coin number, the non-normalization of the size, the interference of other characters and the diversity of the number arrangement, it is difficult to realize the automatic detection, recognition and interpretation of the Euro coin year by using computer vision algorithm. Method The method of detecting and recognizing euro coin year in this paper consists of two steps: first, we use Faster-RCNN to detect the number. The model algorithm is mainly completed in four steps: the first step is to send the whole image to be detected into the convolution neural network to get the convolution feature map; the second step is to input the feature map into the RPN network to get multiple candidate regions of the target; the third step is to use the ROI pooling layer to extract the features of the candidate regions; the fourth step is to use the multi-task classifier to carry out position regression to get the precise position coordinates of the target. A self-built experimental platform was used to collect five large coins from 12 EU countries. The five currencies were 2 euros, 1 euro, 50 euro cents, 20 euro cents and 10 euro cents respectively. In the process of collecting, the coins were rotated at small angles continuously, and the coins were captured at various angles as far as possible. A total of 4429 pictures were collected from different angles; There are four ways to interpret the ranking order of the number of coin years. For a given coin image, which sort of method to interpret the year is the first clear question. According to observation, the year arrangement of a certain currency value in a country is fixed. If we can predetermine the currency value and country of a coin, the corresponding year interpretation rules can be determined. This is a problem that can be solved because the image size of coins with different values varies significantly and the coin patterns of different countries are different. Therefore, the second step is: by using K-means clustering algorithm, the obtained digital candidate boxes can be grouped into 4 categories, and the most confident candidate boxes are selected in each category, finally, according to the pre-determined year arrangement pattern of different country coins, the correct year information can be obtained by proper sorting algorithm. Result On a self-built experimental platform, 4429 pictures were collected from 5 kinds of coins with large currency value in 12 EU countries. The training samples and test samples were divided according to the ratio of 1:1. The experimental results show that the detection accuracy of the method is 89.62%, and the calculation time is about 215ms, which basically meets the requirements of accuracy and real-time. Conclusion The algorithm has good performance of real-time, robustness and high precision, and has high practical application value. Although the detection accuracy of existing algorithms is close to 90%, there is still much room for improvement, which can be considered from two aspects to solve the existing two error situations. One is to improve the clustering algorithm to achieve compact clustering or clustering in accordance with the law of year number distribution, which can eliminate the misdetection of characters or symbols to a certain extent; the other is to further improve the Faster-RCNN network model and the simplified processing algorithm of candidate boxes to improve the detection accuracy of closely arranged digital boxes.
Keywords
QQ在线


订阅号|日报