摘要
钢铁微观组织分析是根据钢铁材料的显微组织特点,对材料的性能进行分析的过程。目前微观组织识别往往依靠专业人员判断,需要大量人力物力,效率低、容易受到主观因素的影响而造成结果的不确定。研究了深度神经网络中基于残差结构的微观组织智能分析问题,通过对残差网络模型的改进,提出了基于迁移学习的改进残差网络模型,在ImageNet数据集上进行预训练,并将权值迁移到改进残差网络模型中,实现小样本数据集下的深度学习。该卷积神经网络模型在16种钢铁材料微观组织测试集上进行了测试,结果表明,该方法的准确率达到95.36%,较基础网络结构识别率提高了6.9个百分点。与其他网络结构模型相比,该模型不仅识别率高而且泛化能力强。
Steel microstructure analysis is the process of analyzing the properties of steel materials according to the microstructure characteristics of steel materials.At present,the identification of micro-organization often relies on the judgment of professionals,which requires a lot of manpower and material resources,and is low in efficiency and easy to be affected by subjective factors,resulting in uncertain results.By improving the residual network model,an improved residual network model based on transfer learning is proposed.The ImageNet data set is pre-trained and the weights are transferred to the improved residual network model to realize deep learning under the small sample data set.The convolutional neural network model is tested on 16 kinds of steel material microstructure test sets,and the results show that the accuracy of the method is 95.36%,which is 6.9 percentage points higher than that of the basic network structure recognition rate.Compared with other network structure models,this model not only has high recognition rate but also strong generalization ability.
作者
宋月
安治国
白丽娟
严文谨
宋召朝
谷秀锐
刘丽君
张青
刘子韬
Song Yue;An Zhiguo;Bai Lijuan;Yan Wenjin;Song Zhaochao;Gu Xiurui;Liu Lijun;Zhang Qing;liu Zitao(HBIS Materials Technology Research Institute,Shijiazhuang 050023,Hebei;HBIS Digital Technology Co.,Ltd.,Shijiazhuang 050023,Hebei)
出处
《河北冶金》
2024年第3期57-60,75,共5页
Hebei Metallurgy
关键词
深度学习
残差网络
权值
迁移学习
微观组织
智能识别
deep learning
residual network
weight
transfer learning
microstructure
intelligent recognition
