基于残差优化和内容自适应的文本识别算法

摘要：

西林瓶标签信息在保障患者用药安全和高效的药物管理方面发挥着关键作用.针对传统的文本识别网络对药瓶标签图像中长文本和模糊文本的识别性能差的间题，提出了一种基于残差优化和内容自适应的文本识别算法.在传统文本识别网络的基础上，采用多尺度残差特征提取模块来代替原有的特征提取卷积网络，通过优化ResNet 网络的下采样过程并引入多尺度特征融合模块，增强了特征提取能力.同时，加入卷积注意力模块提升了网络对文本的关注,增强了网络对低分辨率文本的识别能力.其次，在序列建模阶段，融合多层双向内容自适应递归单元和自注意力机制，提升了长文本序列的建模能力.实验结果表明，与卷积递归神经网络文本识别网络相比，本算法识别准确率提高了3.92%，相较于其他文本识别网络相比均有一定的提升.

The labeling information of penicilin bottles plays a crucial role in ensuring patient medication safety and effi-cient drug management. A text recognition algorithm based on residual optimization and content adaptation is proposed to address the problem of poor recognition performance of traditional text recognition networks for long and fuzzy text in medicine bottle label images. On the basis of traditional text recognition networks, by optimizing the down sampling process of ResNet network and introducing the multi-scale feature fusion module, the ability of feature extraction is enhanced. At the same time.the addition of CBAM attention mechanism has improved the networks attention to the text, a multi-scale residual feature ex-traction module is adopted to replace the original feature extraction convolutional network, enhancing the network's ability to recognize low resolution text. Secondly, in the sequence modeling stage, the fusion of muli-layer bidirectional content adaptive recursive units and self attention mechanisms enhances the modeling capability of long text sequences. The experimental results show that compared with the CRNN text recognition network, this algorithm has improved the recognition accuracy by 3.92%.which is a certain improvement compared to other text recognition networks.

作者: 

王敏，黎永顺，欧翔，曹冉，吴佳

Wang Min,Li Yongshun,Ou Xiang,Cao Ran,Wu Jia

机构地区：

南京信息工程大学电子与信息工程学院；安徽建筑大学电子与信息工程学院,

引用本文：

王敏，黎永顺，欧翔等。基于残差优化和内容自适应的文本识别算法[J].河南师范大学学报(自然科学版)，2026,54(2):30-37. (Wang Min, Li Yongshun, Ou Xiang, et al. Residual optimization and content-adaptive text recognition algorithm[J].Journal of Henan Normal University (Natural Science Edition).2026. 54 (2):30-37.DOI:10.16366/j.cnki.1000-2367.2024.11.02.0002.)

基金：

国家自然科学基金；安徽省高校杰出青年科研项目

关键词：

文本识别；CRNN；特征提取；内容自适应循环单元；自注意力机制

text recognition; CRNN; feature extraction; CARU; self-attention mechanim

分类号：

TP391.4

基于残差优化和内容自适应的文本识别算法.pdf