一种变电站设备表计缺陷图像识别方法
A Novel Image Recognition Method for Substation Meter Device Defects
陈永昕 1杜镇安 2黎恒烜 3张侃君 3姚伟 4龙昌武 5滕捷3
作者信息
- 1. 强电磁工程与新技术国家重点实验室(华中科技大学),湖北 武汉 430074;国网湖北省电力有限公司电力科学研究院,湖北 武汉 430077
- 2. 国网湖北省电力有限公司,湖北 武汉 430048
- 3. 国网湖北省电力有限公司电力科学研究院,湖北 武汉 430077
- 4. 强电磁工程与新技术国家重点实验室(华中科技大学),湖北 武汉 430074
- 5. 国网湖北省电力有限公司黄石供电公司,湖北 黄石 435000
- 折叠
摘要
随着新一代变电站集中监控系统的建设和应用,海量巡视数据的汇集加速了人工智能在设备管控领域的应用,自动快速识别变电站设备缺陷对构建"无人值守+集中监控"变电运维新模式有重要意义.提出了一种基于改进型YOLOv9的识别方法,在YOLOv9的骨干网络层引入了基于跨空间学习的高效多尺度注意力机制(EMA)以提取变电站设备表计图像中小目标的关键特征,并引入了Inner-SIoU边框回归损失函数提高YOLOv9的收敛性和准确性.实验结果表明,该方法的准确率(P)、召回率(R)和平均精度(mAP)均优于基线方法.
Abstract
With the construction and application of a new-generation substation centralized monitoring system,the collection of massive inspection data has accelerated the application of artificial intelligence(AI)in the field of equipment management and control.Automatic and rapid identification of substation equipment defects is of vital importance for the construction of a new mode of"unattended+centralized monitoring"substations'operation and maintenance.Therefore,this paper proposes an improved YOLOv9-based image recognition method:an efficient multi-scale attention(EMA)mechanism based on cross-space learning is introduced into the backbone network layer of YOLOv9 to extract key features of small and medium-sized targets in meter images of substation equipment.Meanwhile,Inner-SIoU bounding box regression loss function is used to improve the convergence and accuracy of YOLOv9.The experimental results show that the precision rate(P),recall rate(R),and mean average precision(mAP)of the proposed method are superior to those of the baseline method.
关键词
变电站/无人值守/集中监控/图像识别/表计/YOLOv9/注意力机制/边界框回归损失函数Key words
substation/unattended/centralized supervisory control/image recognition/meter device/YOLOv9/attention mechanism/bounding box regression loss function引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据