基于改进YOLOv5的光伏电池EL缺陷检测算法
EL defect detection algorithm for photovoltaic cells based on improved YOLOv5
王宗良 1陆丽1
作者信息
- 1. 上海电机学院 电气学院, 上海 201306
- 折叠
摘要
针对光伏电池电致发光(EL)缺陷检测中存在的数据不均衡和缺陷尺度不一等因素导致的误检、漏检问题,提出了一种改进的YOLOv5缺陷检测算法.首先,采用K-means++聚类算法和IoU距离公式产生先验框,避免聚类中心出现局部最优解;其次,将网络中的C3模块替换成梯度流更丰富的C2f模块,减少训练过程中产生的特征图冗余;最后,将原有损失函数CIoU优化为WIoU,加速收敛并提高回归精度.实验结果表明:该模型在光伏电池EL数据集上平均精度均值上达到0.942,相比于原始网络提高了2.6%,能对光伏电池EL缺陷图像进行有效地定位识别,对光伏电池EL缺陷图像检测有实际应用价值.
Abstract
Aiming at the problem of misdetection and omission caused by factors such as data imbalance and defect scale inequality in the detection of electroluminescence defects in photovoltaic cells,an improved YOLOv5 defect detection algorithm is proposed.Firstly,the K-means++ clustering algorithm and the IoU distance formula are used to generate a priori frames to avoid the emergence of local optimal solutions in the clustering centre;secondly,the C3 module in the network is replaced by the C2f module with richer gradient flow to reduce the redundancy of the feature maps generated during the training process;lastly,the original loss function,CIoU,is optimized to WIoU to accelerate the convergence and to improve regression accuracy.The experimental results show that the model reaches 0.942 on the average accuracy mean value on the PV cell EL dataset,which is 2.6%higher than the original network,and can effectively locate and identify the EL defect images of PV cells,which is of practical application for the detection of defective images of photovoltaic cell EL.
关键词
光伏电池/电致发光/缺陷检测Key words
photovoltaic cell/electroluminescence/defect detection引用本文复制引用
出版年
2024
NETL
NSTL
万方数据