| 基于MFBN-YOLOv5输电线路绝缘子缺陷检测研究 |
 点此下载全文 |
| 引用本文：王凯1，黄陈蓉2，顾杰1，季星宇1.基于MFBN-YOLOv5输电线路绝缘子缺陷检测研究[J].计算技术与自动化,2025,(1):80-87 |
| 摘要点击次数: 996 |
| 全文下载次数: 253 |
|
|
| 中文摘要:针对输电线路上绝缘子检测效率低下的问题，提出了一种绝缘子缺陷检测模型MFBN-YOLOv5。首先针对骨干网络中特征提取能力不足的问题，本文设计了MC3模块替换骨干网络中的C3模块，在骨干网络尾部引入了Fenhence模块，模块采用卷积与空洞卷积串联的方式，有效扩大感受野并增强特征提取。其次，为了改进模型特征融合的能力，在颈部引入BiFPN结构，提升网络对绝缘子不同缺陷的特征融合能力。最后，边界回归损失函数使用NWD(Normalized Wasserstein Distance)度量方式，提高了在复杂背景条件下模型对绝缘子缺陷的定位精度。实验结果表明，改进后的MFBN-YOLOv5模型可以快速、准确地检测绝缘子的缺陷，平均精度均值（mAP0.5）达到95.6%，比原YOLOv5s模型高3.9%，能够满足日常电力巡检的需求。 |
| 中文关键词:绝缘子缺陷 MC3 空洞卷积 改进特征融合 NWD |
| |
| Research on Defect Detection of Insulators in Transmission Lines Based on MFBN-YOLOv5 |
|
|
| Abstract:In order to solve the problem of low efficiency of insulator detection on transmission lines, an insulator defect detection model MFBN-YOLOv5 is proposed. Firstly, in order to solve the problem of insufficient feature extraction ability in the backbone network, the MC3 module is designed to replace the C3 module in the backbone network, and the Fenhence module is introduced at the tail of the backbone network. Secondly, in order to improve the feature fusion ability of the model, the BiFPN structure is introduced into the neck to improve the feature fusion ability of the network to different defects of the insulator. Finally, the boundary regression loss function uses the Normalized Wasserstein Distance (NWD) metric to improve the positioning accuracy of the model for insulator defects under complex background conditions. Experimental results show that the improved MFBN-YOLOv5 model can quickly and accurately detect insulator defects, and the average accuracy (mAP0.5) reaches 95.6%, which is 3.9% higher than that of the original YOLOv5s model (mAP0.5), which can meet the needs of daily power inspection. |
| keywords:insulator defects MC3 dilated convolution improved feature fusion NWD |
| 查看全文 查看/发表评论 下载pdf阅读器 |
次