Swin Transformer的电网设备缺陷检测与识别研究
Swin Transformer-based detection and identification of defects in power grid equipment
戴永东 1李明江 1王茂飞 1蒋承伶 2马洲俊2
作者信息
- 1. 国网江苏省电力有限公司泰州供电公司,泰州 225300
- 2. 国网江苏省电力有限公司,南京 210000
- 折叠
摘要
针对传统图像处理算法在解决复杂环境电网设备图像中所出现的识别速度慢、检测精度差等问题,提出了一种基于改进Swin Transformer网络的电网设备缺陷检测与识别方案.首先,利用伪色彩图像增强机制与边缘高频信息图像分割机制对原始图像进行高频特征信息筛选与提取.然后,在Swin Transformer主干网络中引入复合标记特征金字塔模块,解决多尺度语义特征的提取成功率低的问题.最后通过构建弱监督目标选择机制,实现对电网设备图像中目标的检测与分类.实验结果表明,文中所设计的算法在电网设备缺陷与监测上效果显著,识别精度可达92.6％.
Abstract
Traditional image processing algorithms are slow in recognition and poor in detection of the images of grid equipment in complex environments.Based on an improved Swin Transformer network,this paper presents a scheme for the detection and recognition of defects in grid equipment.First,the original images are screened and extracted for high-frequency feature information using pseudo-color image enhancement mechanism and edge high-frequency information image segmentation mechanism.Then,a composite labeled feature pyramid module is introduced into the Swin Transformer backbone network to improve the success probability in extraction of multi-scale semantic features.Finally,a weakly supervised target selection mechanism is built for the detection and classification of targets in grid equipment images.Experimental results show that the algorithm proposed in this paper is effective in monitoring grid equipment and detecting its defects,the recognition accuracy reaching 92.6％.
关键词
图像处理/特征提取/Swin/Transformer/电网设备/目标分类Key words
image processing/feature extraction/Swin Transformer/grid equipment/target classification引用本文复制引用
出版年
2024
NETL
NSTL
万方数据