机械制造与自动化2026，Vol.55Issue(2)：114-120,7.DOI:10.19344/j.cnki.issn1671-5276.2026.02.022
融入反光特征的金属压印字符识别深度学习模型
Deep Learning Model for Metal Embossed Character Recognition Incorporating Reflective Features
摘要
Abstract
To improve the low recognition accuracy of characters embossed on train line nameplates due to the reflection of metal materials,a deep learning model for recognition of metal embossed characters incorporating reflective features is proposed,namely RFF-YOLOv5-C3CA.The proposed model,based on YOLOv5,introduces the reflective feature fusion(RFF)module to enhance the contrast of the imprinted character area in the image in order to extract effective image features,uses target attention module C3CA with direction information and position information reinforcing the model's perception of embossed characters.The experiment results show that compared with the YOLOv5 model,the recognition acurcy,recall rate,and mAP value of the proposed model on the train nameplate embossed character data set increasesoon by 3.46 percentage points,1.41 percentage points,and 1.72 percentage points respectively,and the detection speed FPS reaches 41 f/s,which demonstrates the effectiveness of the proposed algorithm.关键词
压印字符/反光特征/YOLOv5/字符识别/注意力机制Key words
embossed characters/reflective features/YOLOv5/character recognition/attention mechanism分类
信息技术与安全科学引用本文复制引用
赵睿栋,李立,欧阳,崔大宾..融入反光特征的金属压印字符识别深度学习模型[J].机械制造与自动化,2026,55(2):114-120,7.基金项目
四川省重大科技专项项目(2022ZDZX0007) （2022ZDZX0007）
