| 摘 要: 同一热轧钢坯生产线上会存在钢坯表面字符的字体不一致的问题，而利用深度学习YOLOv3算法训练不同字体的字符数据集，严重影响了整体字符的识别率，虽然原始的YOLOv3网络结构适用性较好，但对喷印字符识别区域没有针对性。为解决以上问题，根据喷印字符相对较小且没有大小形态变化的特性，改进了YOLOv3模型结构，仅保留预测小、中目标的网络结构，在保证较高检测精度的同时，缩小模型容量；采用对不同字体字符分开训练的识别方式，得出针对性分开训练比混合字体整体训练的识别准确率高的结论。结果表明，本方法比不同字体整体训练的识别准确率提高了7%以上，可在工程上进行应用。 |
| 关键词: 深度学习；字符识别；热轧钢坯；YOLOv3 |
|
中图分类号: TP301.6
文献标识码: A
|
|
| Research on Character Recognition of Different Fonts on the Surface of Hot Rolled Steel Billet based on Deep Learning |
|
LIU Kang1, QIAN Wei1, YANG Kang2
|
( 1.School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China;
2.Shanghai Baosight Software Co ., Ltd ., Shanghai 201999, China)
1010898612@qq.com; 1458515538@qq.com; yangkang@baosight.com
|
| Abstract: Aiming at character fonts inconsistency on the billet surface in the same hot-rolled billet production line, deep learning YOLOv3 algorithm is used to train character data sets of different fonts, which seriously affects the overall character recognition rate. Although the original YOLOv3 network structure is quite applicable, it is not targeted at the recognition area of printed characters. In order to solve the above problem, this paper proposes to improve YOLOv3 model structure according to the characteristics of relatively small print characters and no changes in size and shape. Only the network structure for predicting small and medium targets is retained, and the model capacity was reduced while ensuring high detection accuracy. It is concluded that the recognition accuracy of the targeted separate training is higher than that of the whole training of mixed fonts. The results show that the recognition accuracy of this method is more than 7% higher than that of the whole training of different fonts, and it can be applied in engineering. |
| Keywords: deep learning; character recognition; hot-rolled steel billet; YOLOv3 |
