摘要
对电路面板的元器件识别,可大量减少人工成本和时间,避免因人为主观因素而产生不合格样品。根据应用场景要求,提出ResNet50融合EfficientNet的骨架网络方法。该方法增强了物体检测的语义信息,有效解决多尺度检测问题,对各类尺度检测问题有着更好的识别效果。通过骨架网络的特征提取后,加载到双向FPN(Bi-Feature Pyramid networks)层再进行更深层次的特征提取,将高语义特征和浅层的定位特征进行融合。实验结果表明,使用该构建的融合模型,具有精确率高、检测图片的帧率高、鲁棒性好的优点,具有良好的泛化能力。
The component identification of circuit panel can greatly reduce labor cost and time,and avoid unqualified samples due to human subjective factors.According to the requirement of application scenario,this paper proposes a skeleton network method of ResNet50 combined with EfficienctNet.This method enhanced the speech information of object detection,effectively solved the problem of multi-scale detection,and had better recognition effect for various scale detection problems.After the feature extraction of skeleton network,it was loaded into the Bi-feature pyramid networks(Bi-FPN)layer,and further feature extraction was carried out to fuse the high semantic features and the shallow location features.The experimental results show that the proposed fusion model has the advantages of high precision,high frame rate,robustness and good generalization ability.
作者
欧阳家斌
胡维平
侯会达
Ouyang Jiabin;Hu Weiping;Hou Huida(School of Electronic Engineering,Guangxi Normal University,Guilin 541000,Guangxi,China;Suzhou Institute of Automation,Chinese Academy of Sciences,Suzhou 215000,Jiangsu,China)
出处
《计算机应用与软件》
北大核心
2023年第5期119-123,共5页
Computer Applications and Software
关键词
电路面板
电子元件识别
深度学习
网络融合
对比实验
Circuit panel
Electronic component identification
Deep learning
Network fusion
Comparative experiment