基于改进YOLOv5的光伏太阳能电池片缺陷检测  

Defect Detection of Photovoltaic Solar Cells Based on Improved YOLOv5

作　　者：郭建 黄颖驹 GUO Jian;HUANG Yingju(School of Mechanical Engineering,Guangzhou City University of Technology,Guangzhou 510800,China)

机构地区：[1]广州城市理工学院机械工程学院,广州510800

出　　处：《组合机床与自动化加工技术》2024年第11期104-109,共6页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：2022广东省普通高校特色创新人才类项目(2022KTSCX185);广州城市理工学院校级青年科研基金项目(K0222005);2023年广东省科技创新战略专项资金项目(pdjh2023b0778);广州城市理工学院2023年度校级科研基金项目(K0223002);2024年广东省科技创新战略专项资金项目(pdjh2024a528)。

摘　　要：针对光伏电池片缺陷检测中存在的样本数据量不均衡、缺陷尺寸差异较大和背景复杂等传统目标检测算法无法解决而导致的误检、漏检问题,提出了一种基于改进的YOLOv5的光伏电池片缺陷检测算法。首先,在数据准备阶段,使用BEGAN生成对抗网络进行数据增强,扩充缺陷图片数据集,处理种类之间不平衡和缺陷尺寸差异问题;其次,在颈部网络中使用BiFPN双向特征金字塔网络,通过提取不同层次的特征信息以融合更多的缺陷特征,从而减少光伏组件复杂背景的干扰,提高检测性能;最后,在模型检测输出层添加小目标检测头,减少微小微弱缺陷信息的丢失,避免特征混淆,提高检测精度。实验结果表明,改进后的检测模型应用于数据增强扩充后的EL缺陷数据集检测,综合性能指标F1达到了84.43%,相较传统的YOLOv5算法准确率和召回率分别提升了3.02%和7.13%,检测精度mAP@0.5提高了4.31%。Aiming at the problems of false detection and missed detection caused by unbalanced sample data,large difference in defect size and complex background in photovoltaic cell defect detection,a photovoltaic cell defect detection algorithm based on improved YOLOv5 is proposed.Firstly,in the data preparation stage,the BEGAN generative adversarial network is used to enhance the data,expand the defect image dataset,and deal with the problems of imbalance and defect size difference between classes.Secondly,the BiFPN bidirectional feature pyramid network is used in the neck network,which extracts different levels of feature information to fuse more defect features,thereby reducing the interference of the complex background of PV modules and improving the detection performance.Finally,a small target detection head is added to the model detection output layer to reduce the loss of small and weak defect information,avoid feature confusion,and improve the detection accuracy.The experimental results show that the improved detection model is applied to the detection of EL defect dataset after data enhancement and enrichment,and the comprehensive performance index F1 reaches 84.43%,which improves the accuracy and recall rate by 3.02%and 7.13%,respectively,and the detection accuracy mAP@0.5 increases by 4.31%compared with the traditional YOLOv5 algorithm.

关 键 词：缺陷检测 YOLOv5 深度学习 数据增强 

分 类 号：TH165[机械工程—机械制造及自动化] TG659[金属学及工艺—金属切削加工及机床]