摘要
为解决板坯喷涂面标实时识别问题,构建文本检测和识别模型。改进可微二值化算法网络,引入高效通道注意力模块SENet,进行自适应空间特征融合(ASFF),增强特征金字塔预测多尺度目标的能力。识别模型改进卷积递归神经网络的VGG网络,将卷积与循环神经网络联合训练。实验结果表明,检测模型的精确率、召回率和调和平均值达到93.30%、86.45%、89.85%,提升显著;识别模型平均准确率达到86.01%,精度提升4.99%。模型满足实时与准确性要求。
To solve the problem of real-time recognition of slab spraying surface,a text detection and recognition model was constructed.The differentiable binary algorithm network was improved,the efficient channel attention module SENet was introduced,and adaptive spatial feature fusion(ASFF)was carried out,which enhanced the ability of predicting multi-scale targets.The VGG network of convolution recurrent neural network was improved in the recognition model,and CNN and RNN were jointly trained.Experimental results show that the accuracy,recall and harmonic average of the detection model are 93.30%,86.45%and 89.85%.The average accuracy of the recognition model is 86.01%,and the accuracy is improved by 4.99%.The model meets the requirements of real-time and accuracy.
作者
董维振
陈燕
梁海玲
DONG Wei-zhen;CHEN Yan;LIANG Hai-ling(School of Computer,Electronics and Information,Guangxi University,Nanning 530004,China;Guangxi China Tobacco Industry Limited Company,Nanning 530001,China)
出处
《计算机工程与设计》
北大核心
2023年第1期116-124,共9页
Computer Engineering and Design
基金
国家自然科学基金项目(31771775)
广西自然科学基金项目(2020GXNSFAA159090)
国家重点研发计划专项基金项目(2017YFB0304002)。
关键词
板坯喷涂面标
可微二值化
高效通道注意力机制
特征金字塔
自适应空间特征融合
卷积递归神经网络
联合训练
反向传播
迁移学习
surface mark of slab spraying
differentiable binarization
efficient channel attention mechanism
feature pyramid networks
adaptive spatial feature fusion
convolutional recurrent neural network
joint training
back propagation
transfer learning
