Scholarlink

<p>Real-time detection of photovoltaic panel defects remains highly challenging, as the model must simultaneously overcome algorithmic performance bottlenecks and background interference. To address these issues, this paper proposes an improved real-time detection framework, CHS-YOLO. The core innovations include: the C3k2_DEConv module, which enhances multi-scale defect perception; the HSPAN neck network, which integrates multi-scale spatial attention with hierarchical feature reorganization; and the Shape-IoU loss, which introduces shape-consistency constraints to optimize bounding-box regression accuracy and shape fitting, thereby significantly improving the detection of tiny cracks and irregular defects. Experiments conducted on a photovoltaic panel defect dataset demonstrate that CHS-YOLO achieves higher detection accuracy than the original YOLOv11, with mAP@0.5 improved by 4.3%, recall improved by 5.3%, and precision improved by 3.7%, while reducing model parameters by 26.9%. Moreover, the model achieves an inference speed of 294.1 FPS, indicating that it simultaneously delivers high accuracy and real-time performance. This study provides an efficient and practical solution for lightweight and real-time photovoltaic panel defect detection.</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

CHS-YOLO: enhanced lightweight YOLOv11 model for accurate photovoltaic panel defect detection

  • 期刊论文

  • 发表日期：

    2026年2月23日

查看全文
  • Zhao Tang,
  • Dongfeng Tang

摘要

Real-time detection of photovoltaic panel defects remains highly challenging, as the model must simultaneously overcome algorithmic performance bottlenecks and background interference. To address these issues, this paper proposes an improved real-time detection framework, CHS-YOLO. The core innovations include: the C3k2_DEConv module, which enhances multi-scale defect perception; the HSPAN neck network, which integrates multi-scale spatial attention with hierarchical feature reorganization; and the Shape-IoU loss, which introduces shape-consistency constraints to optimize bounding-box regression accuracy and shape fitting, thereby significantly improving the detection of tiny cracks and irregular defects. Experiments conducted on a photovoltaic panel defect dataset demonstrate that CHS-YOLO achieves higher detection accuracy than the original YOLOv11, with mAP@0.5 improved by 4.3%, recall improved by 5.3%, and precision improved by 3.7%, while reducing model parameters by 26.9%. Moreover, the model achieves an inference speed of 294.1 FPS, indicating that it simultaneously delivers high accuracy and real-time performance. This study provides an efficient and practical solution for lightweight and real-time photovoltaic panel defect detection.