面向嵌入式平台的智能巡视点位校正系统研究  

Research on Intelligent Patrol Point Correction System for Embedded Platform

作　　者：陈俊杰 洪心皓 CHEN Junjie;HONG Xinhao(State Grid Fujian Electric Power Co.,Ltd.,Fuzhou,China,350100)

机构地区：[1]国网福建省电力有限公司,福州350100

出　　处：《福建电脑》2025年第2期88-92,共5页Journal of Fujian Computer

基　　金：国网福建省电力有限公司群众性创新项目(No.52130A23000L)资助。

摘　　要：变电站的智能巡视存在误检率较高和监控设备智能化程度不足的问题。为解决这些问题,本文提出一种应用于嵌入式设备的智能巡视点位校正系统。系统由软件模块和嵌入式设备构成,软件模块主要由前端控制平台和加注意力机制的YOLOv5模型组成,硬件模块主要完成YOLOv5检测模型在嵌入式设备上的量化编译和硬件加速。实验结果表明:本文方法的推理速度为23帧/s,在以浮标式油位表作为典型待检测目标的50次校正测试中,平均用时为4.5s,误差小于5%,性能能够满足实际变电站巡视点位校正的工程需要。The intelligent inspection of substations has the problems of high false detection rate and insufficient intelligence level of monitoring equipment。To address these issues,this article proposes an intelligent patrol point calibration system for embedded devices.The system is composed of software module and embedded device.The software module is mainly composed of front-end control platform and YOLOv5 model with attention mechanism,while the hardware module mainly completes quantitative compilation and hardware acceleration of YOLOv5 detection model on embedded device.The experimental results show that the inference speed of our method is 23 frames per second;In 50 calibration tests using a float type oil level gauge as a typical target to be tested,the average time was 4.5 seconds and the error was less than 5%.Its performance can meet the engineering needs of actual substation inspection point calibration.

关 键 词：智能巡检 AI前端算法 点位校正 嵌入式设备 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]