苏晨1,2, 任志俊1,2, 范彪1,2, 董俊杰1,2.基于注意力机制与ResNet的残余奥氏体评级研究[J].轻工机械,2023,41(2):78-84 |
基于注意力机制与ResNet的残余奥氏体评级研究 |
Research on Retained Austenite Rating Based on Attention Mechanism and ResNet |
|
DOI:10.3969/j.issn.1005 2895.2023.02.012 |
中文关键词: 残余奥氏体 评级模型 注意力机制 ResNet 迁移学习 |
英文关键词:retained austenite rating model attention mechanism ResNet transfer learning |
基金项目: |
|
摘要点击次数: 251 |
全文下载次数: 595 |
中文摘要: |
针对目前残余奥氏体评级受限于金相设备与研究者的工作经验,不确定因素较多的问题,课题组采用迁移学习与CBAM优化ResNet50模型对残余奥氏体等级进行识别,并构建残余奥氏体级别评级模型,最后使用测试数据集对于模型复杂度与准确度进行验证。实验结果表明:该模型对于残余奥氏体金相图谱识别性较强,等级识别准确率达到941%,并且对于其他金相组织也有较好的泛化能力,能够满足现场检测需求。 |
英文摘要: |
To address the problem of the current residual austenite rating limited by the metallographic equipment and the working experience of the researcher resulting in many uncertainties, migration learning and CBAM was used to optimize the ResNet50 model for retained austenite grade recognition, a residual austenite grade rating model was constructed, and finally a test data set was used to verify the complexity and accuracy of the model. The experimental results show that the model has strong recognition ability for retained austenite metallographic image with an accuracy of 94.1%, and also has good generalization ability for other metallographic structures, which can meet the needs of field inspection. |
查看全文 查看/发表评论 下载PDF阅读器 |
关闭 |