摘要
使用机器学习技术,开发了复合材料结构背光图像缺陷识别系统,实现了对内部缺陷的智能识别。使用迁移学习技术,并通过人工修正的识别结果来更新数据库并修正模型参数,实现了模型的在线学习,提高了模型的准确率。使用PyQt5开发的识别软件的界面,实现了数据标注、缺陷识别、结果修正等功能,并保留了进一步开发的接口。由于机器学习模型存在误差,通过人工修正的识别结果来更新数据库并修正模型参数,实现了模型的在线学习,提高了模型的准确率。同时,将数据标注和结果修正的功能一体化,便于软件的使用和拓展。开发的系统对复合材料结构背光图像中缺陷的识别高效、准确,软件界面易于使用,提高了复合材料结构背光检测的效率和可靠性。
Using machine learning technology,a defect recognition system for backlit images of composite material structure is developed,which realizes the intelligent recognition of internal defects.Using transfer learning technology,the database is updated and the model parameters are corrected by manual correction of the recognition results,which realizes the online learning of the model and improves the accuracy of the model.Using the interface of the recognition software developed by PyQt5,the functions of data annotation,defect recognition,result correction and so on are realized,and the interface for further development is reserved.Because of the error of the machine learning model,the database is updated and the model parameters are corrected by manual correction of the recognition results,which realizes the online learning of the model and improves the accuracy of the model.At the same time,the function of data annotation and result correction is integrated,which is convenient for the use and expansion of the software.The developed system is efficient and accurate in the recognition of defects in the backlit images of composite material structure,and the software interface is easy to use,which improves the efficiency and reliability of the backlight detection of composite material structure.
作者
肖鹏
夏泽宇
张翰轩
XIAO Peng;XIA Zeyu;ZHANG Hanxuan(Composite Materials Center,Shanghai Aircraft Manufacturing Co.,Ltd.,Shanghai 201324,China)
出处
《复合材料科学与工程》
北大核心
2024年第S1期6-10,29,共6页
Composites Science and Engineering
关键词
背光检测
复合材料
智能分析
迁移学习
缺陷识别
backlight detection
composite material
intelligent analysis
transfer learning
defect identification
