基于YOLOV7的改进汽车零件缺陷检测算法
Detection Method of Auto Parts Defects Based on Improved YOLOV7
申科翔 1方群1
作者信息
- 1. 安徽师范大学 计算机与信息学院,安徽 芜湖 241003
- 折叠
摘要
汽车零部件缺陷检测通常采用基于YOLOV7 的深度学习模型,针对生产环境中计算资源受限的问题,提出了改进YOLOV7 的汽车零部件表面缺陷检测算法,首先使用MobileNetV3模块替换YOLOV7骨干网络,其次将颈部网络替换为重复加权双向特征金字塔网络BiFPN和多路径高效设计BepC3模块,最后采用一种基于高斯Wasserstein距离的回归损失函数代替原有的损失函数.实验结果表明,该算法相较传统YOLOV7的AP50精度提高0.155,比YOLOx算法AP50精度提高0.139,在检测精度和效率方面综合表现优异.
Abstract
The defect detection of automotive components usually adopts a deep learning model based on YO-LOV7.In response to the problem of limited computing resources in the production environment,an improved YOLOV7 surface defect detection method for automotive components is proposed.Firstly,the lightweight net-work model MobileNetV3 is used to replace the YOLOV7 backbone feature extraction network.Secondly,the neck network is replaced with a repeated weighted bidirectional feature pyramid network BiFPN and a multi-path efficient design BepC3 module.Finally,a regression loss function based on Gaussian Wasserstein distance is used to replace the original loss function.The experimental results show that the algorithm improves the accuracy of AP50 by 0.155 compared to the traditional YOLOV7 and 0.139 compared to the YOLOx algorithm.It performs excellently in both detection accuracy and efficiency.
关键词
零件缺陷检测/YOLOV7/MobileNetV3/数据增强/轻量化网络Key words
part defect detection/YOLOV7/Mobile NetV3/data enhancement/lightweight network引用本文复制引用
出版年
2024
NETL
NSTL
万方数据