{"title":"基于机器学习的钢组织检测质量水平估计系统","authors":"Hiromi Nishiura, A. Miyamoto, Akira Ito, Shogo Suzuki, Kouhei Fujii, Hiroshi Morifuji, Hiroyuki Takatsuka","doi":"10.1093/jmicro/dfac019","DOIUrl":null,"url":null,"abstract":"For quality control of special steels, the microstructure of the steel is visually inspected on the basis of microscopic images. In this study, aiming to eliminate the effect of personal differences between inspectors and reduce inspection costs, a system for automatically estimating quality level (hereafter, “automatic-quality-level-estimation system ‘’) based on machine learning is proposed and evaluated. Collecting the images is a manual task performed by the inspector, and it is difficult to prepare multiple training samples in advance. As for the proposed method, overfitting, which is a problem in training with few samples, is suppressed by data expansion based on variation distribution of correct-answer values. The correct-answer rate for judging quality level by an inspector was about 90%, while the proposed method achieved a rate of 90%, which is sufficient to render the method practically applicable.","PeriodicalId":48655,"journal":{"name":"Microscopy","volume":"71 1","pages":"214 - 221"},"PeriodicalIF":1.5000,"publicationDate":"2021-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"2","resultStr":"{\"title\":\"Machine-learning-based quality-level-estimation system for inspecting steel microstructures\",\"authors\":\"Hiromi Nishiura, A. Miyamoto, Akira Ito, Shogo Suzuki, Kouhei Fujii, Hiroshi Morifuji, Hiroyuki Takatsuka\",\"doi\":\"10.1093/jmicro/dfac019\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"For quality control of special steels, the microstructure of the steel is visually inspected on the basis of microscopic images. In this study, aiming to eliminate the effect of personal differences between inspectors and reduce inspection costs, a system for automatically estimating quality level (hereafter, “automatic-quality-level-estimation system ‘’) based on machine learning is proposed and evaluated. Collecting the images is a manual task performed by the inspector, and it is difficult to prepare multiple training samples in advance. As for the proposed method, overfitting, which is a problem in training with few samples, is suppressed by data expansion based on variation distribution of correct-answer values. The correct-answer rate for judging quality level by an inspector was about 90%, while the proposed method achieved a rate of 90%, which is sufficient to render the method practically applicable.\",\"PeriodicalId\":48655,\"journal\":{\"name\":\"Microscopy\",\"volume\":\"71 1\",\"pages\":\"214 - 221\"},\"PeriodicalIF\":1.5000,\"publicationDate\":\"2021-07-25\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"2\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Microscopy\",\"FirstCategoryId\":\"5\",\"ListUrlMain\":\"https://doi.org/10.1093/jmicro/dfac019\",\"RegionNum\":4,\"RegionCategory\":\"工程技术\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q3\",\"JCRName\":\"MICROSCOPY\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Microscopy","FirstCategoryId":"5","ListUrlMain":"https://doi.org/10.1093/jmicro/dfac019","RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"MICROSCOPY","Score":null,"Total":0}
Machine-learning-based quality-level-estimation system for inspecting steel microstructures
For quality control of special steels, the microstructure of the steel is visually inspected on the basis of microscopic images. In this study, aiming to eliminate the effect of personal differences between inspectors and reduce inspection costs, a system for automatically estimating quality level (hereafter, “automatic-quality-level-estimation system ‘’) based on machine learning is proposed and evaluated. Collecting the images is a manual task performed by the inspector, and it is difficult to prepare multiple training samples in advance. As for the proposed method, overfitting, which is a problem in training with few samples, is suppressed by data expansion based on variation distribution of correct-answer values. The correct-answer rate for judging quality level by an inspector was about 90%, while the proposed method achieved a rate of 90%, which is sufficient to render the method practically applicable.
期刊介绍:
Microscopy, previously Journal of Electron Microscopy, promotes research combined with any type of microscopy techniques, applied in life and material sciences. Microscopy is the official journal of the Japanese Society of Microscopy.