基于I-YOLO模型的轴瓦零件缺陷视觉检测方法

Visual Detection Method for Defects in Bearing Parts Based on I-YOLO Model

吴晓成 ¹缪晨炜 ¹王要波 ¹陈鹏¹

扫码查看

作者信息

1. 中国船舶重工集团公司第七○三研究所,江苏无锡 214151
折叠

摘要

轴瓦是水力测功器的重要零部件,其质量影响机组的安全运行.基于传统机器视觉的轴瓦缺陷检测方法存在准确率低、漏检率高的问题.针对该问题,提出一种基于改进YOLO模型的轴瓦缺陷检测方法.首先,对数据集进行图像预处理,扩大训练样本;然后,改进YOLO模型结构,通过候选框维度聚类和多尺度训练,提出一种新的I-YOLO深度学习模型;最后,在轴瓦数据集训练和测试I-YOLO模型,并将该模型与YOLOv4、YOLOv3 模型进行对比.结果表明,改进的I-YOLO模型在轴瓦缺陷检测中的准确率达到了 98.73%,比YOLOv4、YOLOv3 模型分别提高了3.01%、10.88%,证明改进的模型能在提高检测准确率的同时有效降低漏检率.

Abstract

Bearing is an important component of hydraulic dynamometer,and its quality affects the safe operation of the unit.The traditional machine vision based bearing defect detection method has the problems of low accuracy and high missed detection rate.A bearing defect detection method based on an improved YOLO model is proposed to address this issue.Firstly,image preprocessing is performed on the dataset to expand the training samples.Then,by improving the YOLO model structure,a new I-YOLO deep learning model is proposed through candidate box dimension clustering and multi-scale training.Finally,train and test the I-YOLO model on the bearing dataset,and compare it with the YOLOv4 and YOLOv3 models.The result shows that the improved I-YOLO model achieves an accuracy of 98.73%in bearing defect detection,which is 3.01%and 10.88%higher than the YOLOv4 and YOLOv3 models,respectively.This proves that the improved model can effectively reduce the missed detection rate while improving detection accuracy.

关键词

深度学习/轴瓦/卷积神经模型/视觉识别/缺陷检测

Key words

deep learning/bearings/convolutional neural models/visual recognition/defect detection

引用本文复制引用

出版年

2024

自动化应用

重庆西南信息有限公司

自动化应用

影响因子：0.156

ISSN：1674-778X

参考文献量9

段落导航