目的 针对目前印刷套准识别方法依赖于经验人工设计特征提取的问题，提出一种不需要人工提取图像特征的卷积神经网络模型，实现印刷套准状态的识别。方法 采用图像增强技术实现不均衡训练集的均衡化，增加训练集图像的数量，提高模型的识别准确率。设计基于AlexNet网络结构的印刷套准识别模型的结构参数，分析批处理样本数量和基础学习率对模型性能的影响规律。结果 文中方法获得的总印刷套准识别准确率为0.9860，召回率为1.0000，分类准确率几何平均数为0.9869。结论 文中方法能自动提取图像特征，不依赖于人工设计的特征提取方法。在构造的数据集上，文中方法的分类性能优于实验中的支持向量机方法。

The current printing registration recognition methods usually use the feature extraction of experienced manual design. To solve this problem, a convolutional neural network model without manual image feature collection is proposed to realize printing registration recognition. The image enhancement technology is used to equalize the imbalanced training set to increase the amount of training set and improve the recognition accuracy of the model. The structural parameters of the printing registration recognition model based on AlexNet network are designed, and the effects of batch sample number and basic learning rate on the model performance are analyzed. The proposed method achieves promising experimental results. The total accuracy of printing registration recognition is 0.9860 with the recall of 1.0000 and the geometric means of classification accuracy of 0.9869. The method in this paper automatically extracts image features, and does not rely on artificially designed feature extraction methods. On the constructed data set, the classification performance of the proposed method is superior to the experimental support vector machine method.