为了实现易拉罐喷码过程中罐底字符的检测,提出了一种基于机器视觉的实时检测方法。首先对采集的图像进行二维Arimoto熵阈值化处理,然后在进行形态学膨胀操作后,通过字符区域信息提取出喷码字符区域,并进行旋转校正。将校正后的图像分割为单个字符,归一化为28×28大小,并采用卷积神经网络分类器进行检测识别。试验表明,该算法鲁棒性好,可以适应不同型号的易拉罐图像的分割识别,且实时性和准确性高,可满足易拉罐点阵喷码字符在线检测的要求。

To achieve the detection of canning’s characters in the process of printing,a realtime detection method based on machine vision is proposed. Firstly,the 2DArimoto entropy is processed on collected images,the morphological expansion operation is carried out,the characters’ region by character regional information and the rotation correction is done. Besides,the adjusted images are segmented into single character images,28×28 sizes are normalized to and the detection is done by adopting Convolutional Neural Network classifier. Experiments demonstrate that the algorithm has good robustness,strong adaptability to segment and recognize different types of cannings,high instantaneity and accuracy,and it can meet the demand of online detection of canning dotmatrix printed characters.