摘要
针对采用人工方法对芯片进行分选和重排时的精度、速度和鲁棒性都达不到工业现场要求的情况,提出了一种基于改进YOLOv4的半导体芯片表面字符识别方法。该算法通过轻量化YOLOv4的特征提取网络,保留更多的浅层特征信息;在非特征提取网络引入深度可分离卷积,减少模型参数;在网络头部嵌入卷积注意力模块(BAM),进一步提升对小尺寸字符的特征提取能力。通过对工业半导体芯片的识别实验结果表明,与原YOLOv4相比,改进的方法具有更高的识别精度和更快的识别速度,且具有良好的鲁棒性和泛化能力,能够满足工业现场字符识别的需求。
Aiming at the situation that the accuracy,speed and robustness of chip sorting and rearrangement by manual method can not meet the requirements of industrial field,a surface character recognition method of semiconductor chip based on improved YOLOv4 is proposed.The algorithm preserves more shallow feature information by lightweight YOLOv4 feature extraction network;The depth separable convolution is introduced into the non feature extraction network to reduce the model parameters;CBAM module is embedded in the network header to further improve the feature extraction ability of small-size characters.The experimental results of industrial semiconductor chip recognition show that compared with the original YOLOv4,the improved method has higher recognition accuracy and faster recognition speed,and has good robustness and generalization ability,which can meet the needs of industrial field character recognition.
作者
安胜彪
娄慧儒
白宇
An Shengbiao;Lou Huiru;Bai Yu(School of Information Science and Engineering,Hebei University of Science and Technology,Shijiazhuang,050018,China)
出处
《国外电子测量技术》
北大核心
2022年第4期77-82,共6页
Foreign Electronic Measurement Technology
关键词
深度学习
字符识别
YOLOv4
半导体芯片
deep learning
character recognition
YOLOv4
industrial scene
semiconductor chip
