基于YOLOv5的太阳电池表面缺陷检测
RESEARCH ON SURFACE DEFECT DETECTION OF SOLAR CELL WITH IMPROVED YOLOv5 ALGORITHM
彭自然 1张颖清 1肖伸平1
作者信息
- 1. 湖南工业大学电气与信息工程学院,株洲 412007
- 折叠
摘要
针对太阳电池表面缺陷问题,在深度学习模型YOLOv5的基础上进行优化与改进.首先,为充分利用深层、浅层和原始的特征信息,加强特征融合,设计具有跨连接结构的特征金字塔网络(ScFPN).其次,为加强多重感受野融合,基于SPPF构建SPPFCSPC模块,通过最大池化层获得不同感受野,提升算法对于不同尺度太阳电池缺陷检测的鲁棒性.最后,采用ASD-IoU作为边界框损失函数,提升边框回归的速度与精度.实验结果表明,改进后的YOLOv5模型mAP@(0.50～0.95)达到83.1%,相比于YOLOv5模型,平均精度提高3.3个百分点,表明该文模型更加适合于太阳电池表面缺陷检测.
Abstract
To solve the surface defect problem of solar cells,the deep learning model YOLOv5 is optimized and improved.Firstly,in order to make full use of deep,shallow and original feature information and strengthen feature fusion,a feature pyramid network(ScFPN)with cross-connection structure is designed.Secondly,in order to strengthen the fusion of multiple receptive fields,the SPPFCSPC module is constructed based on SPPF,and different receptive fields are obtained through the maximum pooling layer,which improves the robustness of the algorithm for the defect detection of solar cells of different scales.Finally,ASD-IoU is used as the bounding loss function to improve the speed and precision of the bounding regression.The experimental results show that the improved YOLOv5 model mAP@(0.50-0.95)reaches 83.1%,and the average accuracy is increased by 3.3 percentage points compared with the YOLOv5 model,indicating that this model is more suitable for surface defect detection of solar cells.
关键词
深度学习/太阳电池/缺陷/卷积神经网络/目标检测/图像处理Key words
deep learning/solar cell/defect/convolutional neural network/object detection/image processing引用本文复制引用
基金项目
湖南省教育厅重点科研项目(22A0423)
湖南省自然科学基金(2023JJ602672022JJ50073)
出版年
2024
NETL
NSTL
万方数据