基于视频识别的气象观测场设备监控技术研究  

Research on Equipment Monitoring Technology of Meteorological Observation Field based on Video Recognition

作　　者：魏春梅 马尚昌[1] 卢会国[1] 黄胃建 WEI Chunmei;MA Shangchang;LU Huiguo;HUANG Weijian(College of Electronic Engineering,Chengdu University of Information Technology,Chengdu 610225,China)

机构地区：[1]成都信息工程大学电子工程学院,四川成都610225

出　　处：《成都信息工程大学学报》2023年第2期129-135,共7页Journal of Chengdu University of Information Technology

基　　金：国家自然科学基金资助项目(42075129)。

摘　　要：针对观测场长期无人监管,存在设备被人破坏和偷盗,且目前还没有对观测场设备进行实时监控的有效算法和系统等问题,比较目前典型的目标检测算法,将其应用于观测场设备的检测和识别。根据研究目标搭建数据获取平台,研究数据预处理算法,比较经典目标检测算法SSD和Yolov5两种模型在气象观测设备识别中的效果。两种模型实验对比结果表明,SSD的识别准确率为92.09%,但训练模型速度较快;Yolov5的识别准确率为95.82%,收敛很快,且识别结果较佳。In view of the long-term unsupervised observation field,the equipment is damaged and stolen,and there is no effective algorithm and system for real-time monitoring of observation field equipment,the current typical target detection algorithms are compared and applied to the detection and identification of observation field equipment.The data acquisition platform is built according to the research objectives,the data preprocessing algorithm is studied,and the effects of two classical target detection algorithms SSD and Yolov5 model in the recognition of meteorological observation equipment are compared.The experimental results show that the detection speed of SSD is faster,but the accuracy is only 92.09%,the accuracy of Yolov5 is 95.82%and the convergence is fast.

关 键 词：目标检测 SSD Yolov5 数据增强 视频识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]