An improved U-Net algorithm for extracting carbides from spray formed high speed steel

doi:10.13251/j.issn.0254-6051.2023.10.012

Abstract

Abstract: To address the issues of low accuracy, poor extraction effect, and large analysis errors caused by using digital image processing algorithms and deep learning models for quantitative analysis of carbides in steel, a modified jet forming high speed steel carbides segmentation algorithm based on U-Net (GSG-Unet) was proposed. The aim is to accurately and efficiently segment and extract different types of carbides in steel for automated quantitative analysis. The model is strengthened by adding ConvNext module and CBAM attention mechanism to enhance its feature extraction capability and ability to handle missed detections, resulting in significant improvement in segmentation performance. The results show that the improved model has an accuracy of 91.31%, recall rate of 87.52%, class-average intersection over union of 84.89%, and Dice of 83.16%, which are significantly higher than those of the original model. This improved model can accurately segment MC and M₆C carbides from martensite matrix, providing strong technical support for rapid and accurate automated quantitative analysis of carbides in high speed steel.

Key words: high speed steel, carbides, automated quantitative analysis, deep learning, semantic segmentation

CLC Number:

TG113

Chen Jiashu, Hou Guodong, Zhou Jikuan, Liu Tianqi, Deng Baichuan, Zhang Xianglin. An improved U-Net algorithm for extracting carbides from spray formed high speed steel[J]. Heat Treatment of Metals, 2023, 48(10): 87-93.

References

[1]Tekumalla Sravya, Tosi Riccardo, Tan Xipeng, et al. Directed energy deposition and characterization of high-speed steels with high vanadium content[J]. Additive Manufacturing Letters, 2022, 2: 100029.
[2]彭翰林. 精冲模具用粉末冶金高速钢S390/S790的热处理规律研究[D]. 武汉: 华中科技大学, 2020.
[3]赵顺利. 喷射成形高合金高速钢的组织与性能研究[D]. 上海: 上海大学, 2017.
[4]王桂棠, 邓耀华, 吴黎明. 基于数字图像处理技术的金相组织定量分析[J]. 金属热处理, 2006, 31(2): 66-70.
Wang Guitang, Deng Yaohua, Wu Liming. Metallographic quantitative analysis using digital image process[J]. Heat Treatment of Metals, 2006, 31(2): 66-70.
[5]Hecht M D, Webler B A, Picard Y N. Digital image analysis to quantify carbide networks in ultrahigh carbon steels[J]. Materials Characterization, 2016, 117: 134-143.
[6]Smeets S, Ångström J, Olsson C O A. Quantitative phase analysis for carbide characterization in steel using automated electron diffraction[J]Steel Research International, 2019, 90(1): 1-7.
[7]张吉. 基于面积直方图的碳化物不均匀度评级方法[D]. 广州: 广东工业大学, 2010.
[8]单陇红. 基于数字图像的GCr15轴承钢中碳化物均匀性分析[D]. 兰州: 兰州理工大学, 2016.
[9]Khan M Z, Gajendran M K, Lee Y, et al. Deep neural architectures for medical image semantic segmentation: Review[J]. IEEE Access, 2021, 9: 83002-83024.
[10]Yu Hongshan, Yang Zhengeng, Tan Lei, et al. Methods and datasets on semantic segmentation: A review[J]. Neurocomputing, 2018, 304: 82-103.
[11]Liu Z, Mao H, Wu C, et al. A ConvNet for the 2020s[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2022: 11966-11976.
[12]Vaswani A, Shazeer N M, Parmar N, et al. Attention is all you need[C]//Thirty-first Conference on Neural Information Processing Systems. 2017.
[13]Xie S, Girshick R B, Dollár P, et al. Aggregated residual transformations for deep neural networks[C]//2017IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2017: 5987-5995.
[14]Sandler M, Howard A G, Zhu M, et al. MobileNetV2: Inverted residuals and linear bottlenecks[C]//2018IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2018: 4510-4520.
[15]Zhao Z X, Chen K X, Yamane S. CBAM-Unet++: Easier to find the target with the attention module "CBAM"[C]//2021 IEEE 10th Global Conference on Consumer Electronics (GCCE). 2021: 655-657.
[16]柴铮, 汪嘉业, 赵春晖, 等. 面向工业监控典型监督任务的深度迁移学习方法: 现状、挑战与展望[J]. 中国科学: 信息科学, 2023, 53(5): 821-840.