CLCRNet: An Optical Character Recognition Network for Cigarette Laser Code
Wei Zhou, Li Zheng, Xiangchengzhen Li, Z.-r. Yang, Jun Yi · IEEE Transactions on Instrumentation and Measurement · 2024
Automatic recognition of laser codes on cigarette case is a crucial method for distinguishing the authenticity of cigarettes. However, automatic recognition of cigarette laser code is a challenging industrial problem due to the complex background of the cigarette case. In this paper, an optical character recognition method based on convolutional recurrent neural network is proposed for cigarette laser code recognition, named as CLCRNet. To begin with, the CX-block module is developed to replace the convolution module in the feature extraction stage, thus extracting rich semantic information from cigarette laser code. Furthermore, the DW-block module is designed in the feature fusion stage, which adopts convolutional downsampling to better preserve detailed semantic information and improve the accuracy of the network. In addition, we utilize the 2 : 2 : 2 : 2 stage compute ratio to achieve a trade-off between accuracy and speed. Finally, we propose a classification-based contrastive learning method that accelerates the convergence speed during training process, making the model more stable without affecting its inference speed. To validate the effectiveness of the proposed CLCRNet, extensive experiments are validated on the real cigarette laser code dataset. The proposed algorithm not only improves the detection accuracy of cigarette laser code, but also reduces the parameters of the network. Experimental results demonstrate that the proposed method achieves better performance than state-of-the-art character recognition models.