Scholarlink

<p>Ensuring the quality of photovoltaic cells is paramount for enhancing the efficiency of solar energy systems. Traditional defect detection methods struggle with feature extraction and suffer from low accuracy in identifying surface defects. To tackle these challenges, we propose YOLOv8-DG, an enhanced YOLOv8 model tailored for defect detection in electroluminescence images of photovoltaic cells. Firstly, YOLOv8-DG integrates Adaptive Channel Conv and Adaptive Channel Combination Spatial Pyramid Pooling Fast in the backbone to boost computational efficiency. Secondly, it replaces standard convolutions with Channel Ghost Shuffle Conv and C2f with Omni-Dimensional Dynamic Convolution to enable effective channel interaction. Finally, a robust Real-Time DEtection Transformer detection head is employed to enhance detection performance. Several electroluminescent photovoltaic defect datasets are used to verify the effectiveness of the proposed model. The experimental results show that the map@50 and map@50:95 of YOLOv8-DG are 89.5% and 61.4%, which are 7.1% and 5.9% higher than the original YOLOv8, respectively. These findings highlight the potential of YOLOv8-DG in accurately detecting small defects, contributing to improved quality control in photovoltaic manufacturing. The source code is available at <a href="https://github.com/snmae/YOLOv8-DG">https://github.com/snmae/YOLOv8-DG</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing defect detection in photovoltaic cells: a dynamic group YOLOv8 approach

  • 期刊论文

  • 发表日期：

    2025年12月9日

查看全文
  • Huhao Shen,
  • Xin Shu,
  • Xiaofang Guo,
  • Changbin Shao,
  • Zhibin Xie

摘要

Ensuring the quality of photovoltaic cells is paramount for enhancing the efficiency of solar energy systems. Traditional defect detection methods struggle with feature extraction and suffer from low accuracy in identifying surface defects. To tackle these challenges, we propose YOLOv8-DG, an enhanced YOLOv8 model tailored for defect detection in electroluminescence images of photovoltaic cells. Firstly, YOLOv8-DG integrates Adaptive Channel Conv and Adaptive Channel Combination Spatial Pyramid Pooling Fast in the backbone to boost computational efficiency. Secondly, it replaces standard convolutions with Channel Ghost Shuffle Conv and C2f with Omni-Dimensional Dynamic Convolution to enable effective channel interaction. Finally, a robust Real-Time DEtection Transformer detection head is employed to enhance detection performance. Several electroluminescent photovoltaic defect datasets are used to verify the effectiveness of the proposed model. The experimental results show that the map@50 and map@50:95 of YOLOv8-DG are 89.5% and 61.4%, which are 7.1% and 5.9% higher than the original YOLOv8, respectively. These findings highlight the potential of YOLOv8-DG in accurately detecting small defects, contributing to improved quality control in photovoltaic manufacturing. The source code is available at https://github.com/snmae/YOLOv8-DG.