基于机器学习的铝熔体夹渣自动检测技术  

Automatic Detection Technology for Inclusion in Al Melt Based on Machine Learning

作　　者：白蕊 胡勇[1] 金泽发 刘宏泉 闫志杰 Bai Rui;Hu Yong;Jin Zefa;Liu Hongquan;Yan Zhijie(School of Materials Science and Engineering,Taiyuan University of Science and Technology,Taiyuan 030024;School of Materials and New Energy,Ningxia University,Yinchuan 750021;Dongliang Aluminium Co.,Ltd.,Huzhou 313000;School of Materials Science and Engineering,North University of China,Taiyuan 03005l;Shanxi Key Laboratory of Advanced Metal Materials for Special Environments,Taiyuan 030051)

机构地区：[1]太原科技大学材料科学与工程学院,太原030024 [2]宁夏大学材料与新能源学院,银川750021 [3]栋梁铝业有限公司,湖州313000 [4]中北大学材料科学与工程学院,太原030051 [5]特殊环境先进金属材料山西省重点实验室,太原030051

出　　处：《特种铸造及有色合金》2024年第4期477-484,共8页Special Casting & Nonferrous Alloys

基　　金：山西省自然科学基金资助项目(202103021224279);山西省科技成果转化引导专项资助项目(202204021301025);山西省科技创新人才团队资助项目(202204051002020)。

摘　　要：根据现有夹渣图像的特点,提出基于YOLOv5模型的夹渣目标检测算法,以减少获取图像的角度、光源等不确定因素对测试结果所造成的负反馈影响,提高检测精度。利用Mosaic数据增强、自适应锚框计算、自适应图像缩放等技术,融合Focus和CSP结构,设计出基于YOLOv5的自动化识别夹渣图像和自动计算夹渣率的优化算法。结果表明,相对于人工采集照片计算夹渣率水平的方法,改进后的YOLOv5s模型,有效提高了断面夹渣图像目标检测的精确度,由改进前的83%提高至97%。An optimized target detection algorithm based on YOLOv5 model was proposed according to characteristics of existing slag inclusion images to reduce the negative feedback effects caused by uncertain factors such as angle and light source, improving the accuracy. According to Mosaic data enhancement, adaptive anchor frame calculation, adaptive image scaling and other technologies, an algorithm for automatic recognition of slag inclusion images and automatic calculation of slag inclusion rate was designed combined with Focus and CSP structures. The results indicate that the improved YOLOv5s model can effectively enhance the target detection accuracy of sectional slag inclusion images from 83% to 97%, compared with the manually collecting method to calculate slag inclusion rate level.

关 键 词：熔体质量检测 夹渣 YOLOv5 目标检测 

分 类 号：TG27[金属学及工艺—铸造] TG146.21[一般工业技术—材料科学与工程] TU512.4[金属学及工艺—金属材料]