Scholarlink

<p>Insulator condition monitoring is of great significance for ensuring the safe operation of power systems. However, in actual inspection scenarios, insulator defects (such as break, drop, and flashover) often exhibit characteristics such as minute sizes, complex background interference, and large-scale differences, causing existing detection algorithms to be prone to missed detections and poor real-time performance. To address these challenges, this paper proposes a novel insulator defect detection model named SIDF-YOLO (Small Insulator Defect Focus YOLO). First, a CLS module is designed to replace the original C3k2 module. By synergizing Large Kernel Perception with Small Kernel Dynamic Aggregation for feature extraction, it enhances the model’s ability to represent minute defects in complex backgrounds. Second, a Small Target Focus Pyramid Network (STFPN) is proposed, which adopts SPDConv to achieve lossless downsampling, effectively preserving fine-grained spatial information. Finally, by introducing a lightweight detection head, the model balances speed with detection accuracy. Experimental results demonstrate that SIDF-YOLO achieves an mAP50 of 88.1% and an mAP50:95 of 57.9%. Crucially, after deployment and acceleration on the Rockchip RK3588 edge computing platform, the model achieves an actual inference speed of 39 FPS, effectively meeting the performance requirements for real-time UAV inspection of power lines. This study provides a reliable solution for high-precision insulator defect detection in resource-constrained environments.</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

SIDF-YOLO: focusing on small targets for real-time insulator defect detection

  • 期刊论文

  • 发表日期：

    2026年7月18日

查看全文
  • Defu Chen,
  • Zihao Jian,
  • Sheng Xiang,
  • Mingye Li,
  • Xianbao Wang

摘要

Insulator condition monitoring is of great significance for ensuring the safe operation of power systems. However, in actual inspection scenarios, insulator defects (such as break, drop, and flashover) often exhibit characteristics such as minute sizes, complex background interference, and large-scale differences, causing existing detection algorithms to be prone to missed detections and poor real-time performance. To address these challenges, this paper proposes a novel insulator defect detection model named SIDF-YOLO (Small Insulator Defect Focus YOLO). First, a CLS module is designed to replace the original C3k2 module. By synergizing Large Kernel Perception with Small Kernel Dynamic Aggregation for feature extraction, it enhances the model’s ability to represent minute defects in complex backgrounds. Second, a Small Target Focus Pyramid Network (STFPN) is proposed, which adopts SPDConv to achieve lossless downsampling, effectively preserving fine-grained spatial information. Finally, by introducing a lightweight detection head, the model balances speed with detection accuracy. Experimental results demonstrate that SIDF-YOLO achieves an mAP50 of 88.1% and an mAP50:95 of 57.9%. Crucially, after deployment and acceleration on the Rockchip RK3588 edge computing platform, the model achieves an actual inference speed of 39 FPS, effectively meeting the performance requirements for real-time UAV inspection of power lines. This study provides a reliable solution for high-precision insulator defect detection in resource-constrained environments.