Scholarlink

<p>Photovoltaic cell crack detection is critical for maintaining the efficiency and reliability of solar energy systems. However, existing detection algorithms often struggle with the trade-off between model size and accuracy, particularly in identifying complex crack patterns in photovoltaic cells. To address this challenge, we propose PSD-YOLO, a novel detection algorithm designed to optimize both performance and efficiency. The algorithm integrates the lightweight convolution module PSDConv into YOLOv7-tiny, replacing the DW convolution in Ghost Spatial Convolution (GSConv), thereby enhancing the detection capability for diverse crack types while retaining adaptability. Additionally, the loss function is enhanced by adopting the Efficient IoU (EIoU) in place of the original CIoU, which contributes to faster convergence and improved bounding box regression accuracy. Experimental results demonstrate that PSD-YOLO reduces parameters and computational cost by 18.3% and 16.7%, respectively, compared to YOLOv7-tiny, resulting in a more lightweight and resource-efficient model. However, this improvement in compactness comes with a trade-off in inference speed, which has been noted in our experimental results.</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

YOLOv7-tiny-based lightweight and efficient algorithm for photovoltaic cell crack detection

  • 期刊论文

  • 发表日期：

    2025年6月9日

查看全文
  • Satyarth Tiwari,
  • Sanjay Kumar Sharma

摘要

Photovoltaic cell crack detection is critical for maintaining the efficiency and reliability of solar energy systems. However, existing detection algorithms often struggle with the trade-off between model size and accuracy, particularly in identifying complex crack patterns in photovoltaic cells. To address this challenge, we propose PSD-YOLO, a novel detection algorithm designed to optimize both performance and efficiency. The algorithm integrates the lightweight convolution module PSDConv into YOLOv7-tiny, replacing the DW convolution in Ghost Spatial Convolution (GSConv), thereby enhancing the detection capability for diverse crack types while retaining adaptability. Additionally, the loss function is enhanced by adopting the Efficient IoU (EIoU) in place of the original CIoU, which contributes to faster convergence and improved bounding box regression accuracy. Experimental results demonstrate that PSD-YOLO reduces parameters and computational cost by 18.3% and 16.7%, respectively, compared to YOLOv7-tiny, resulting in a more lightweight and resource-efficient model. However, this improvement in compactness comes with a trade-off in inference speed, which has been noted in our experimental results.