面向电能表检定流水线的轻量化目标检测算法  

Lightweight Object Detection Algorithm for Electric Meter Calibration Line

作　　者：董贤光 孙艳玲 代燕杰 邢宇 翟晓卉 孙凯 吕玉超 吴强[2] 刘琚[2] DONG Xianguang;SUN Yanling;DAI Yanjie;XING Yu;ZHAI Xiaohui;SUN Kai;LYU Yuchao;WU Qiang;LIU Ju(Marketing Service Center(Metering Center),State Grid Shandong Electric Power Co.Ltd.,Ji’nan 250001,China;School of Information Science and Engineering,Shandong University,Qingdao 266237,China)

机构地区：[1]国网山东省电力公司营销服务中心(计量中心),济南250001 [2]山东大学信息科学与工程学院,青岛266237

出　　处：《数据采集与处理》2025年第2期545-560,共16页Journal of Data Acquisition and Processing

基　　金：国网山东省电力公司科技项目(520633230003)。

摘　　要：在工业流水线场景中,利用带有视觉信息的目标检测技术为故障发现及消缺提供决策信息已成为智能生产的新热点。针对电能表流水线检定场景中目标遮挡、小目标密集排列等问题,在YOLOv8n的基础上,提出了一种轻量化目标检测算法。通过引入O⁃GELAN模块,在保持低计算量的同时获取更丰富的特征层次。利用特征收集⁃融合⁃分发的颈部架构和通道位置注意力机制实现特征跨层融合;此外,采用重参数化卷积检测头以进一步提高检测效率。在现场采集流水线数据上的实验表明,改进后模型的mAP(0.5)和mAP(0.5∶0.95)分别达到了0.994和0.828,检测速度可达111.5帧/s,能够满足工业场景下的高精度和高实时性需要,可为故障消缺提供辅助决策。In the industrial production line scenario,target detection technology with visual information has become a new hotspot for intelligent production,providing decision⁃making information for fault detection and elimination.In response to issues such as target occlusion and dense arrangement of small targets in the electric energy meter production line inspection scenario,this study proposes a lightweight target detection algorithm based on YOLOv8n.By introducing the O⁃GELAN module,the algorithm achieves richer feature levels while maintaining low computational complexity.The neck architecture of feature collection,fusion,and distribution,along with the channel position attention mechanism,enables cross⁃layer feature fusion.Furthermore,a re⁃parameterized convolutional detection head is employed to enhance detection efficiency.Experiments conducted on field⁃collected production line data demonstrate that the improved model’s mAP(0.5)and mAP(0.5∶0.95)have reached 0.994 and 0.828,respectively,with a detection speed of up to 111.5 frames per second.This meets the high precision and real⁃time requirements of industrial scenarios and can provide auxiliary decision⁃making for fault elimination.

关 键 词：工业流水线 密集目标检测 YOLOv8 特征融合 现场数据采集 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]