基于YOLOv7-tiny的化工反应分液面检测研究  

Study on Separate Liquid Level Detection of Chemical Reaction Based on YOLOv7-tiny

作　　者：曹梦旋 陈秀生[1] 袁涛 黄元凤 林心园 李树锐 CAO Meng-xuan;CHEN Xiu-sheng;YUAN Tao;HUANG Yuan-feng;LIN Xin-yuan;LI Shu-rui(School of Mechanical Engineering,University of finan;Shandong Special Inspection Technology Co.,Ltd.)

机构地区：[1]济南大学机械工程学院 [2]山东特检科技有限公司

出　　处：《化工自动化及仪表》2025年第2期205-211,共7页Control and Instruments in Chemical Industry

摘　　要：为实现化工反应分液面快速、高精度的目标检测和提取,设计基于机器视觉的化工反应分液面检测系统,系统由人机交互模块、图像采集装置(光源、工业相机、光学镜头)、电源模块、传输模块、4G模块组成,利用YOLOv7-tiny轻量级目标检测算法进行分液面目标检测识别。在相同实验条件下,选用同一物料数据集进行训练,实验结果表明:YOLOv7-tiny算法的mAP值相对最高,相较于常用的轻量级YOLOv3-tiny和YOLOv5s算法分别高出9.5%和5.1%;YOLOv7-tiny的检测速度较YOLOv3-tiny和YOLOv5s算法分别低5 FPS/s和高10 FPS/s;YOLOv7-tiny算法的内存占用量较YOLOv3-tiny和YOLOv5s算法分别低4.92 MB和低2.07 MB,说明YOLOv7-tiny算法的分类精度和运行速度更具优势。With the view to realizing rapid and high-precision target detection and extraction of the liquid level in chemical reaction,a chemical reaction level detection system based on machine vision was designed.The system,composed of human-computer interaction module,image acquisition device(light source,industrial camera and optical lens),power supply module,transmission module and 4G module,has YOLOv7-tiny lightweight objection detection algorithm adopted to detect and identify liquid level targets,including under the same experimental conditions,having the same material dataset selected for training.The experimental results show that,the mAP value of YOLOv7-tiny algorithm is the highest relatively,which is 9.5%and 5.1%higher than that of the commonly-used lightweight YOLOv3-tiny and YOLOv5s algorithms,respectively.The detection speed of YOLOv7-tiny is 5 FPS/s lower than that of YOLOv3-tiny and 10 FPS/s higher than that of YOLOv5s algorithms.The memory consumption of YOLOv7-tiny algorithm is 4.92 MB and 2.07 MB lower than that of YOLOv3-tiny and YOLOv5s algorithms,respectively.It indicates that the YOLOv7-tiny algorithm has superior advantages in classification accuracy and running speed.

关 键 词：分液面 轻量级目标检测算法 分类精度 运行速度 

分 类 号：TH816[机械工程—仪器科学与技术]