摘要
金相组织分析是钢铁材料研发过程中一项重要的分析手段,目前主要通过经验丰富的专家进行人工判别,费时且容易受到主观意识的影响。为此,研究了基于残差神经网络结构的金相组织智能分析方法,通过对残差网络模型进行改进,提出了基于迁移学习的改进残差网络模型以及基于注意力机制的深度残差收缩网络模型,采用两种不同的卷积神经网络模型在20种钢铁材料微观组织测试集上进行验证,实验结果表明:两种模型的准确率分别达到95.36%和95.79%,泛化能力强,最短平均预测时间仅为1.66 s/张。两种模型在钢铁材料金相组织特征分类方面具有一定的先进性,实现了金相组织类型分类的自动化和智能化。
Microstructure analysis is an important analysis method in the research and development process of iron and steel materials.At present,it is mainly judged manually by experts with rich experience,which is time-consuming and easily affected by subjective consciousness.Therefore,an intelligent analysis method for microstructure based on the residual neural network structure is studied.By improving the residual network model,an improved residual network model based on transfer learning and a deep residual shrinkage network model based on the attention mechanism are proposed.Two different convolutional neural network models are used for verification on the microstructure test set of 20 steel materials.The experimental results show that the accuracies of the two models reach 95.36%and 95.79%respectively,with strong generalization ability,and the shortest average prediction time is only 1.66 s per image.The two models have certain advantages in the classification of microstructure features of iron and steel materials,realizing the automation and intelligence of microstructure type classification.
作者
宋月
严文谨
刘培培
安治国
林亚团
宋召朝
SONG Yue;YAN Wenjin;LIU Peipei;AN Zhiguo;LIN Yatuan;SONG Zhaochao(HBIS Material Technology Research Institute Co.,Ltd.,Shijiazhuang 050023,China;HBIS Digital Technology Co.,Ltd.,Shijiazhuang 050023,China)
出处
《轧钢》
北大核心
2025年第4期111-118,共8页
Steel Rolling
基金
河北省自然科学基金项目(E2020318125)。
关键词
金相组织分析
残差网络
迁移学习
注意力机制
自动化和智能化
microstructure analysis
residual network
transfer learning
attention mechanism
automation and intelligence
