[1]范 涛,王明泉,张俊生,等.基于轻量化 YOLOv4 的轮毂内部缺陷检测算法[J].机械与电子,2023,41(02):3-7.
　FAN Tao,WANG Mingquan,ZHANG Junsheng,et al.Internal Defect Detection Algorithm of Wheel Hub Based on Lightweight YOLOv4[J].Machinery & Electronics,2023,41(02):3-7.
点击复制

基于轻量化 YOLOv4 的轮毂内部缺陷检测算法()
分享到：

《机械与电子》[ISSN:1001-2257/CN:52-1052/TH]

卷:
41
期数:
2023年02期
页码:
3-7
栏目:
设计与研究
出版日期:
2023-02-28

文章信息/Info

Title:
Internal Defect Detection Algorithm of Wheel Hub Based on Lightweight YOLOv4
文章编号:
1001-2257 ( 2023 ) 02-0003-05
作者:
范 涛 1 ; 王明泉 1 ; 张俊生 2 ; 曹鹏娟 1 ; 朱榕榕 1
1. 中北大学仪器科学与动态测试教育部重点实验室,山西 太原 030051 ;
2. 太原工业学院电子工程系,山西 太原 030008
Author(s):
FAN Tao1 ;  WANG Mingquan1 ;  ZHANG Junsheng2 ;  CAO Pengjuan1 ;  ZHU Rongrong1
( 1.Key Laboratory of Instrumentation Science and Dynamic Measurement of Ministry of Education , North University of China , Taiyuan 030051 , China ; 2.Department of Electronic Engineering , Taiyuan Institute of Technology , Taiyuan 030008 , China )
关键词:
目标检测;  YOLOv4 ;  MobileNetV3 ; 轻量化网络; 通道注意力机制
Keywords:
target detection ;  YOLOv4 ;  MobileNetV3 ;  lightweight network ;  SE
分类号:
TP391.4
文献标志码:
A
摘要:
为解决常规深度学习方法检测轮毂内部缺陷存在模型尺寸大、参数多和精度低等问题,提出一种轻量化 YOLOv4 的轮毂内部缺陷检测算法。该算法采用 MobileNetV3 替换 YOLOv4 的主干特征提取网络,并利用深度可分离卷积模块对 YOLOv4 的 PANet ( path aggregation network )模块中的传统卷积进行了替换。同时,在 PANet 特征加强网络中加入通道注意力机制( SE )模块,提高了轮毂内部缺陷目标的识别精度。测试结果表明,所提算法检测精度为 90.23% ,权值文件为 45.2 MB ,检测速率为 68.38 帧/ s 。相较于常规模型性能有所提升,更适用于轮毂内部缺陷的快速、准确检测。
Abstract:
In order to solve the problems of large model size , many parameters and low accuracy in the detection of internal defects of the wheel hub by the conventional deep learning method , this paper proposes a lightweight YOLOv4 wheel internal defect detection algorithm.The algorithm uses MobileNetV3 to replace the backbone feature extraction network of YOLOv4 , and uses the depthwise separable convolution module to replace the traditional convolution in the PANet( path aggregation network ) module of YOLOv4.At the same time , the channel attention mechanism( squeeze and excitation , SE ) module is added to the PANet feature enhancement network , which improves the recognition accuracy of defective targets inside the hub.After a large amount of data testing , the detection accuracy of the algorithm in this study is 90.23% , the weight file is 45.2 MB , and the detection rate is 68.38 frames per second.Compared with the conventional model , the performance has been improved , and it is more suitable for fast and accurate detection of internal defects of the wheel hub.

参考文献/References:

[ 1 ] LI W , LI K S , HUANG Y , et al.Defects of wheel hubs detection and recognition based on trend peak algorithm[ J ] .International journal of embedded systems , 2017 , 9( 3 ): 211-219.

[ 2 ] MAIRE E , WITHERS P J.Quantitative X-ray tomography [ J ] .International materials reviews , 2017 , 59( 1 ): 1-43.
[ 3 ] LIN J H , YU Y , LIN M , et al.Detection of a casting defect tracked by deep convolution neural network [ J ] . International journal of advanced manufacturing technology , 2018 , 97 ( 4 ):573-581.
[ 4 ] FERGUSON M , LEE Y T T , NARAYANAN A , et al. A standardized PMML format for representing convolutional neural networks with application to defect detection [ J ] .Smart and sustainable manufacturing systems , 2019 , 3 ( 1 ): 79-97.
[ 5 ] MERY D.Aluminum casting inspection using deep object detection methods and simulated ellipsoidal defects [ J ] . Machine vision and applications , 2021 , 32 ( 3 ): 1-16.
[ 6 ] 蔡彪,沈宽,付金磊,等 . 基于 Mask R-CNN 的铸件 X 射线 DR 图像缺陷检测研究[ J ] . 仪器仪表学报, 2020 ,41 ( 3 ): 61-69.
[ 7 ] 王陶然,王明泉,张俊生,等 . 基于 Mask R-CNN 的轮毂缺陷分割技术[ J ] . 国外电子测量技术, 2021 , 40 ( 2 ):1-5.
[ 8 ] JIANG L L , WANG Y X , TANG Z H , et al.Casting defect detection in X-ray images using convolutional neural networks and attention-guided data augmentation [ J ] .Measurement , 2021 , 170 : 1-8.
[ 9 ] SANDLER M , HOWARD A , ZHU M L , et al.MobileNetV2 : Inverted residuals and linear bottlenecks [ C ] ∥2018 IEEE / CVF Conference on Computer Vision and Pattern Recognition.Salt Lake : IEEE , 2018 : 4510-4520.
[ 10 ] HOWARD A , SANDLER M , CHU G , et al.Searching for MobileNetV3 [ C ]// 2019 IEEE / CVF International Conference on Computer Vision ( ICCV ) .Seoul : IEEE , 2019 : 1314-1324.
[ 11 ] WANG C Y , LIAO H Y M , WU Y H , et al.CSPNet : A new backbone that can enhance learning capability of CNN [ C ] ∥2020 IEEE / CVF Conference On Computer Vision And Pattern Recognition Workshops( CVPRW ) .New York : IEEE , 2020 : 1571-1580.
[ 12 ] LIU S , QI L , QIN H F , et al.Path aggregation network for instance segmentation [ C ] ∥2018 IEEE / CVF Conference on Computer Vision and Pattern Recognition( CVPR ) .New York : IEEE , 2018 : 8759-8768.
[ 13 ] 赵鹤,杨晓洪,杨奇,等 . 融合注意力机制的金属缺陷图像分割方法 [ J ] . 光电子 · 激光, 2021 ,32( 4 ):403-408.
[ 14 ] 万浪,凌毓涛,郑锡聪,等 . 基于 MobileNetV3-YOLOv4的车型识别[ J ] . 软件导刊, 2021 , 20 ( 12 ): 173-178.

相似文献/References:

[1]卞越洋,高晓科,张伟军.机器人带电作业中的视觉定位与优化策略[J].机械与电子,2021,(05):57.
　BIAN Yueyang,GAO Xiaoke,ZHANG Weijun.Vision Positioning and Optimization Strategyin Robot Live Work[J].Machinery & Electronics,2021,(02):57.
[2]王琪璇,管声启,胡璐萍.基于改进 YOLOv4 的工业棒料识别算法[J].机械与电子,2022,(01):25.
　WANG Qixuan,GUAN Shengqi,HU Luping,et al.Industrial Bar Recognition Algorithm Based on Improved YOLOv4[J].Machinery & Electronics,2022,(02):25.
[3]刘瑞昊,于振中,孙 强.改进多尺度特征融合的工业现场目标检测算法[J].机械与电子,2022,(11):40.
　LIU Ruihao,YU Zhenzhong,SUN Qiang.Improved Multi-scale Feature Fusion for Industrial Field Object Detection Algorithm[J].Machinery & Electronics,2022,(02):40.
[4]张晋钊,刘新妹,等.改进 YOLOv5s 的 PCB 元器件检测技术[J].机械与电子,2024,42(10):35.
　ZHANG Jinzhao,LIU Xinmei,et al.Improved YOLOv5s Component Detection Technology for PCB[J].Machinery & Electronics,2024,42(02):35.

备注/Memo

备注/Memo:
收稿日期: 2022-08-03
基金项目:山西省重点研发计划( 201803D121069 );山西省高等学校科技创新项目( 2020L0624 );山西省信息探测与处理重点实验室基金( ISPT2020 5 )
作者简介:范 涛 ( 1994- ),男,山西朔州人,硕士研究生,研究方向为图像处理;王明泉 ( 1970- ),男,山西朔州人,博士,教授,研究方向为图像处理。
更新日期/Last Update: 2023-03-06