基于深度学习和增强现实的智能变电站仪表读数识别研究

孙蓉蓉; 王程斯; 罗育林; 田松林; 庄秋乃; 夏成文

孙蓉蓉,王程斯,罗育林,田松林,庄秋乃,夏成文.基于深度学习和增强现实的智能变电站仪表读数识别研究[J].电测与仪表,2026,63(5):184-192.
SUN Rongrong,WANG Chengsi,LUO Yulin,TIAN Songlin,ZHUANG Qiunai,XIA Chengwen.Research on intelligent substation instrument reading recognition based on deep learning and augmented reality[J].Electrical Measurement & Instrumentation,2026,63(5):184-192.

基于深度学习和增强现实的智能变电站仪表读数识别研究

Research on intelligent substation instrument reading recognition based on deep learning and augmented reality

DOI：

中文关键词: 智能变电站电力仪表增强现实 YOLOv8模型 DeepLabV3+模型 Transformer模型

英文关键词:intelligent substation, electric power meter, augmented reality, YOLOv8 model, DeepLabV3+ model, Transformer model

基金项目:南网科技项目（090000KK52210151）

作者中文名	作者英文名	单位
孙蓉蓉	SUN Rongrong	深圳供电局有限公司
王程斯	WANG Chengsi	深圳供电局有限公司
罗育林	LUO Yulin	南方电网数字平台科技广东有限公司
田松林	TIAN Songlin	南方电网数字平台科技广东有限公司
庄秋乃	ZHUANG Qiunai	南方电网数字平台科技广东有限公司
夏成文	XIA Chengwen	南方电网数字平台科技广东有限公司

摘要点击次数: 70

中文摘要:

针对现有智能变电站仪表读数识别方法中存在的识别效果不佳和仅能单一识别指针仪表或数字仪表的问题,基于增强现实的电力巡检系统,提出了一种结合改进YOLOv8模型、改进DeepLabV3+模型和改进Transformer模型的智能变电站仪表读数识别方法。改进YOLOv8模型完成仪表分类和区域定位,改进DeepLabV3+模型完成指针式仪表的读取识别,改进Transformer模型完成数字仪表读数识别,通过实验对其性能进行验证。结果表明,改进YOLOv8模型在仪表分类和定位中有效提高了检测精度,检测准确率大于98.00%。改进DeepLabV3+模型在指针式仪表读数识别中有效提高了分割精度,识别误差小于1.50%。改进Transformer模型在数字仪表读数识别中有效提高了识别精度,识别准确率大于97.00%。

英文摘要:

Addressing the issues of poor recognition performance and the ability to only recognize pointer or digital instruments in existing intelligent substation instrument reading recognition methods, based on the augmented reality power inspection system, a smart substation instrument reading recognition method combining improved YOLOv8 model, improved DeepLabV3+model, and improved Transformer model has been proposed. Improve the YOLOv8 model to complete instrument classification and regional positioning, improve the DeepLabV3+model to complete reading and recognition of pointer instruments, and improve the Transformer model to complete digital instrument reading recognition, verify its performance through experiments. The results indicate that, improved YOLOv8 model effectively improves detection accuracy in instrument classification and positioning, with a detection accuracy rate greater than 98.00%. The improved DeepLabV3+model effectively improves segmentation accuracy in pointer instrument reading recognition, with a recognition error of less than 1.50%. The improved Transformer model effectively improves the recognition accuracy in digital instrument reading recognition, with a recognition accuracy rate greater than 97.00%. which can provide certain assistance for the safe operation of the power grid.

查看全文查看/发表评论下载PDF阅读器

关闭