Scholarlink

<p>Ensuring operational reliability in wind energy systems necessitates advanced defect detection methods, yet existing approaches struggle with computational efficiency and sensitivity to subtle anomalies under real-world constraints. This study presents CV-YOLOv8, an optimized framework that enhances defect detection through three methodological advancements: a dilated convolution-based C2f-FocalNextBlock to amplify receptive fields for identifying small and subtle damages on equipment, a hybrid downsampling module integrating strided convolution and max pooling to mitigate feature redundancy, and a multi-scale fusion mechanism to enhance robustness against environmental variations. Evaluations on the DTU Wind Turbine Damage Dataset demonstrate that the proposed architecture achieves an 83.0% mean average precision, representing a 6.1% improvement over baseline YOLOv8n models, while reducing parameter counts by 18.8% (2.6&#xa0;M) and computational complexity to 6.0 GFLOPS. The system maintains practical inference speeds of 66.7 frames per second, confirming its deployability in resource-constrained edge environments. These results establish CV-YOLOv8 as a computationally efficient solution for real-time industrial defect detection, addressing critical gaps in balancing accuracy, model complexity, and operational adaptability for wind turbine maintenance applications. The work underscores the potential of architectural optimization in enabling high-precision, edge-compatible vision systems for sustainable energy infrastructure. The work underscores the potential of architectural optimization in enabling high-precision, edge-compatible vision systems for sustainable energy infrastructure. Furthermore, our approach emphasizes system robustness for real-world monitoring applications, providing a practical solution for real-time 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

CV-YOLOv8: a lightweight and high-accuracy model for wind turbine damage detection

  • 期刊论文

  • 发表日期：

    2025年8月21日

查看全文
  • Li Tuo,
  • Zhao Zhanfang

摘要

Ensuring operational reliability in wind energy systems necessitates advanced defect detection methods, yet existing approaches struggle with computational efficiency and sensitivity to subtle anomalies under real-world constraints. This study presents CV-YOLOv8, an optimized framework that enhances defect detection through three methodological advancements: a dilated convolution-based C2f-FocalNextBlock to amplify receptive fields for identifying small and subtle damages on equipment, a hybrid downsampling module integrating strided convolution and max pooling to mitigate feature redundancy, and a multi-scale fusion mechanism to enhance robustness against environmental variations. Evaluations on the DTU Wind Turbine Damage Dataset demonstrate that the proposed architecture achieves an 83.0% mean average precision, representing a 6.1% improvement over baseline YOLOv8n models, while reducing parameter counts by 18.8% (2.6 M) and computational complexity to 6.0 GFLOPS. The system maintains practical inference speeds of 66.7 frames per second, confirming its deployability in resource-constrained edge environments. These results establish CV-YOLOv8 as a computationally efficient solution for real-time industrial defect detection, addressing critical gaps in balancing accuracy, model complexity, and operational adaptability for wind turbine maintenance applications. The work underscores the potential of architectural optimization in enabling high-precision, edge-compatible vision systems for sustainable energy infrastructure. The work underscores the potential of architectural optimization in enabling high-precision, edge-compatible vision systems for sustainable energy infrastructure. Furthermore, our approach emphasizes system robustness for real-world monitoring applications, providing a practical solution for real-time defect detection.