摘要
由于电梯曳引钢带的工况环境恶劣,钢带表面容易出现缺陷,而复杂昏暗的环境导致缺陷识别和定位困难。为此,提出了基于决策级融合的曳引钢带缺陷检测方法,利用迁移学习的方法微调YOLOv4及SSD的预训练模型,并将其应用于曳引钢带缺陷检测,得到不同的原始检测结果,利用所提方法在决策级融合YOLOv4及SSD模型的原始检测结果,以此提高钢带表面缺陷识别和定位的准确率。利用东北大学钢带表面缺陷公开数据集对所提方法的有效性和可靠性进行测试,缺陷识别和定位结果表明所提方法能够充分利用YOLOv4和SSD模型的原始检测结果,通过决策级融合初始检测结果取得较高的缺陷识别和定位准确性,准确率和召回率提高约15%,交并比达到0.6左右,这对于电梯曳引钢带的缺陷检测和维修保养有重要意义,并进一步保证电梯能够安全可靠运行。
Due to the harsh working environment of the elevator traction steel strip,defects are prone to appear on the surface of the steel strip.In addition,the complex and dim environment makes it difficult to identify and locate the surface defect.Therefore,a defect detection method for traction steel strips is proposed based on decision level fusion.The pre-trained models of YOLOv4 and SSD are fine-tuned using transfer learning,which are applied to defect detection of traction steel strips to obtain different original detection results.The proposed method is used to fuse the original detection results of YOLOv4 and SSD models at the decision level,thereby improving the accuracy of surface defect recognition and localization of steel strips.The effectiveness and reliability of the proposed method are tested using the public dataset of surface defects in steel strips from Northeastern University.The defect identification and localization results show that the proposed method can fully utilize the original detection results of YOLOv4 and SSD models.By fusing the original detection results at the decision level,the high accuracy is achieved for defect identification and localization,where the accuracy and recall increase of about 15%,and the handover to merger ratio of about 0.6.It is of great significance for defect detection and maintenance of elevator traction steel strips,and further ensures the safe and reliable operation of the elevator.
作者
雷高阳
王凯旋
李俊杰
李根生
李海超
Lei Gaoyang;Wang Kaixuan;Li Junjie;Li Gensheng;Li Haichao(School of Information and Control Engineering,China University of Mining and Technology,Xuzhou,Jiangsu 221000,China;School of Information Engineering,Xuzhou College of Industrial Technology,Xuzhou,Jiangsu 221140,China;Xin’an Coal Mine of Henan Dayou Energy Co.,Ltd.,Luoyang,Henan 471842,China;School of Vehicle and Transportation Engineering,Henan University of Science and Technology,Luoyang,Henan 471000,China;Hangzhou Olida Elevator Co.,Ltd.,Hangzhou 311600,China)
出处
《机电工程技术》
2024年第6期205-208,共4页
Mechanical & Electrical Engineering Technology
关键词
曳引钢带
表面缺陷
缺陷检测
决策级融合
图像识别
traction steel strip
surface defects
defect detection
decision-level fusion
image recognition
