Scholarlink

<p>Research on identifying faulty insulators on distribution grids is a primary concern in the research community as it plays a crucial role in maintaining and servicing the electricity supply infrastructure for the public. In this paper, we propose the FGS model to enhance the ability to recognize faulty insulators by combining three methods: (1) improving the Feature Pyramid Network (FPN) to localize insulators in complex background images; (2) introducing a new lightweight network architecture supporting object detection called the Generalized Efficient Layer Aggregation Network (GELAN); (3) incorporating a diffusion model that allows inference based on context while maintaining linear scalability along the sequence length. Additionally, we collected insulator data on utility poles using unmanned aerial vehicles to classify and detect insulator faults, naming this dataset VNelectric. We conducted experiments on the VNelectric dataset, showing that our model achieved 92.3%, 91.4%, and 93.7% precision, recall, and mAP50 respectively.</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

Insulator defect detection based on feature pyramid network and diffusion model

  • 期刊论文

  • 发表日期：

    2025年3月6日

查看全文
  • Anh Trinh Hien,
  • Anh Dat Tran,
  • Dung Cu Viet,
  • Quynh Dao Thi Thuy,
  • Quynh Nguyen Huu

摘要

Research on identifying faulty insulators on distribution grids is a primary concern in the research community as it plays a crucial role in maintaining and servicing the electricity supply infrastructure for the public. In this paper, we propose the FGS model to enhance the ability to recognize faulty insulators by combining three methods: (1) improving the Feature Pyramid Network (FPN) to localize insulators in complex background images; (2) introducing a new lightweight network architecture supporting object detection called the Generalized Efficient Layer Aggregation Network (GELAN); (3) incorporating a diffusion model that allows inference based on context while maintaining linear scalability along the sequence length. Additionally, we collected insulator data on utility poles using unmanned aerial vehicles to classify and detect insulator faults, naming this dataset VNelectric. We conducted experiments on the VNelectric dataset, showing that our model achieved 92.3%, 91.4%, and 93.7% precision, recall, and mAP50 respectively.