Scholarlink

<p>To maximize power generation and enhance energy conversion efficiency, wind turbine blades have been increasingly scaled up. As the primary component responsible for capturing wind energy, these blades are particularly vulnerable to damage under harsh environmental conditions. Additionally, due to the remote locations, expansive areas, and unmanned operations of wind farms, regular inspections are crucial to maintaining safe operation. This paper presents a lightweight small object detection algorithm (LSOD-YOLO) based on YOLOv8, designed for detecting surface damage on wind turbine blades using drone aerial imagery. To tackle the challenge of detecting small objects on wind turbine surfaces, LSOD-YOLO incorporates Omni-dimensional Dynamic Convolution (ODConv) into the C2f module. The neck network is subsequently improved with the Scale Sequence Feature Fusion (SSFF) module and the Triple Feature Encoder (TFE) module. Furthermore, a small object detection layer is introduced to capture additional shallow feature information. These refinements enhance the algorithm’s capacity to detect small objects while preserving accuracy for other target sizes. To achieve a lightweight model design, a strategy involving parameter sharing and partial convolution is employed to optimize the detection head structure. This approach significantly reduces computational load while preserving accuracy. Experimental results on the wind turbine surface damage dataset demonstrate that the proposed LSOD-YOLO algorithm surpasses the baseline in both detection accuracy and model size, facilitating low-latency real-time inference with a notable performance enhancement.</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

LSOD-YOLO: Lightweight Small Object Detection Algorithm for Wind Turbine Surface Damage Detection

  • 期刊论文

  • 发表日期：

    2025年9月1日

查看全文
  • Huanyu Jiang,
  • Hongbing Liu,
  • Zhixiang Chen,
  • Jiufan Hou,
  • Jiajun Liu,
  • Zhenyu Mao,
  • Xianqiang Qu

摘要

To maximize power generation and enhance energy conversion efficiency, wind turbine blades have been increasingly scaled up. As the primary component responsible for capturing wind energy, these blades are particularly vulnerable to damage under harsh environmental conditions. Additionally, due to the remote locations, expansive areas, and unmanned operations of wind farms, regular inspections are crucial to maintaining safe operation. This paper presents a lightweight small object detection algorithm (LSOD-YOLO) based on YOLOv8, designed for detecting surface damage on wind turbine blades using drone aerial imagery. To tackle the challenge of detecting small objects on wind turbine surfaces, LSOD-YOLO incorporates Omni-dimensional Dynamic Convolution (ODConv) into the C2f module. The neck network is subsequently improved with the Scale Sequence Feature Fusion (SSFF) module and the Triple Feature Encoder (TFE) module. Furthermore, a small object detection layer is introduced to capture additional shallow feature information. These refinements enhance the algorithm’s capacity to detect small objects while preserving accuracy for other target sizes. To achieve a lightweight model design, a strategy involving parameter sharing and partial convolution is employed to optimize the detection head structure. This approach significantly reduces computational load while preserving accuracy. Experimental results on the wind turbine surface damage dataset demonstrate that the proposed LSOD-YOLO algorithm surpasses the baseline in both detection accuracy and model size, facilitating low-latency real-time inference with a notable performance enhancement.