Scholarlink

<p>The growing adoption of solar energy underscores the need for efficient defect detection in photovoltaic (PV) modules, as manufacturing flaws and environmental stressors (e.g., cracks, corrosion, humidity) degrade performance and longevity. Hence, it is deemed necessary to identify these defects effectively. Electroluminescence (EL) imaging is an established technique for identifying critical defects like microcracks. Manual inspection remains laborious and error-prone. To address this, we propose a hybrid lightweight deep learning (DL) framework that integrates transformer architectures and attention mechanisms—including DETR, Biformer, and CBAM into the YOLOv8 and YOLOv11 frameworks. This integration enhances feature representation and contextual reasoning while preserving computational efficiency. By automating defect detection through AI-driven computer vision techniques, our approach reduces reliance on manual intervention and improves robustness under real-world conditions. The proposed YOLOv11+CBAM model achieves 93.0% mAP@0.5 with an inference time of 29.3 ms per image and 2,590,815 parameters compared to YOLOv8n with 3,011,823 parameters. Therefore, the YOLOv11+CBAM model outperforms baseline architectures and is suitable for robust real-time photovoltaic cell damage detection using EL imaging.</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

Lightweight YOLOv11 framework with attention mechanisms for robust photovoltaic cell damage detection using electroluminescence imaging

  • 期刊论文

  • 发表日期：

    2026年5月9日

查看全文
  • Harshita Pandey,
  • Sunita Dhavale

摘要

The growing adoption of solar energy underscores the need for efficient defect detection in photovoltaic (PV) modules, as manufacturing flaws and environmental stressors (e.g., cracks, corrosion, humidity) degrade performance and longevity. Hence, it is deemed necessary to identify these defects effectively. Electroluminescence (EL) imaging is an established technique for identifying critical defects like microcracks. Manual inspection remains laborious and error-prone. To address this, we propose a hybrid lightweight deep learning (DL) framework that integrates transformer architectures and attention mechanisms—including DETR, Biformer, and CBAM into the YOLOv8 and YOLOv11 frameworks. This integration enhances feature representation and contextual reasoning while preserving computational efficiency. By automating defect detection through AI-driven computer vision techniques, our approach reduces reliance on manual intervention and improves robustness under real-world conditions. The proposed YOLOv11+CBAM model achieves 93.0% mAP@0.5 with an inference time of 29.3 ms per image and 2,590,815 parameters compared to YOLOv8n with 3,011,823 parameters. Therefore, the YOLOv11+CBAM model outperforms baseline architectures and is suitable for robust real-time photovoltaic cell damage detection using EL imaging.