基于深度学习的轮胎压印字符检测与识别

在自动化装配线上轮胎与轮毂装配时, 需要检测与识别轮胎表面的压印字符串, 从而得到轮胎的品牌、型号、尺寸以及生产的年周号等信息, 用以管理轮胎信息以及监控轮胎的流向。针对轮胎表面压印字符串的检测与识别问题, 文章提出一种基于深度学习的轮胎胎面关键字符的检测与识别方法, 搭建了可编程逻辑控制器(programmable logic controller, PLC)、工控机和工业相机的自动化检测与识别平台, 通过霍夫变换及坐标变换对采集后的图像进行预处理, 采用改进的更快速的区域卷积神经网络(faster region-based convolutional neural network, Faster RCNN)算法为基础检测出目标字符串位置, 再通过卷积递归神经网络(convolutional recurrent neural network, CRNN)对检测出的目标字符串进行识别, 同时利用编码规则校验识别结果, 以提升识别结果的准确率。实验结果表明, 改进后的算法在进行轮胎压印字符串的检测与识别时其准确率超过97.0%, 满足工业生产应用需求。

When assembling tires and wheels on an automated assembly line, it is necessary to detect and identify the imprint character strings on the tire surface, so as to obtain the tire brand, model, size, and production year and week number, which can be used to manage the tire information and monitor the flow direction of tire. Aiming at the detection and recognition of imprint character strings on the tire surface, a method for detecting and recognizing key characters on the tire tread based on deep learning is proposed. An automated detection and recognition platform for programmable logic controller (PLC), industrial computer, and industrial camera is built. The collected images are preprocessed through Hough transform and coordinate transformation, and then the target character string position is detected based on the improved faster region-based convolutional neural network (Faster RCNN) algorithm, and then the detected target character string is recognized by convolutional recurrent neural network (CRNN), and the coding rules are used to verify the recognition result to improve the accuracy of the recognition result. Experimental results show that the improved algorithm has an accuracy rate of over 97.0% when detecting and recognizing tire imprint character strings, which meets the needs of industrial production applications.