基于改进YOLOv5s的电网异物检测算法
Grid Foreign Matter Detection Algorithm Based on Improved YOLOv5s
肖俊阳 1李远 2苏适 3谢青洋3
作者信息
- 1. 中国南方电网深圳供电局有限公司,广东深圳 518000
- 2. 云南电网有限责任公司电力科学研究院,云南昆明 650217;武汉大学电气与自动化学院,湖北武汉 430072
- 3. 云南电网有限责任公司电力科学研究院,云南昆明 650217
- 折叠
摘要
针对电网及沿线异物检测中存在的异物尺度变化、实时性低以及复杂环境下识别精度不足等问题,提出1种基于改进YOLOv5s框架的电网异物检测算法.该方法在主干网络中嵌入ECA注意力机制以减轻背景干扰;同时,采用SPD-Conv模块替换主干网络的卷积模块,引入改进的BiFPN,增强模型对于不同尺寸目标的检测能力.最后,采用Alpha_GIoU损失函数替代原始YOLOv5s中的CIoU部分.经过实验验证,改进YOLOv5s在电网异物检测数据集上mAP的值达到96.98%,能够满足复杂环境下电网异物检测的高效率和高精度要求.
Abstract
Aiming at the problems of foreign object scale change,low real-time performance and insufficient recognition accuracy in complex environment existed in the detection of foreign matter in power grid and along the transmission line,a power grid foreign body detection algorithm based on improved YOLOv5s framework is proposed.ECA attention mechanism is embedded in the backbone network to reduce background interference,SPD-Conv module is used to replace the convolution module of backbone network.Improved BiFPN is introduced to enhance the detection ability of the model for different size targets.Finally,the Alpha_GIoU loss function is used to replace the CIoU part in the original YOLOv5s.The experimental results show that the mAP value of improved YOLOv5s on the grid foreign object detection dataset reaches 96.98%,achieves the grid foreign matter detection demands under complex environment.
关键词
YOLOv5s/电网异物/ECA注意力机制/BiFPN/SPD-Conv/Alpha_GIoUKey words
YOLOv5s/grid foreign matter/ECA attention mechanism/BiFPN/SPD-Conv/Alpha_GIoU引用本文复制引用
基金项目
国家自然科学基金资助项目(51977156)
云南电网公司自筹科技项目(YNKJXM20220207)
出版年
2024
NETL
NSTL
万方数据