基于边缘计算的移栽作业质量监测方法研究  

Research on quality monitoring method of transplanting operation based on edge computing

作　　者：虞飞宇 蒋平 梁佳 Yu Feiyu;Jjiang Ping;Liang Jia(College of Science&Technology Ningbo University,Cixi 315300,Ningbo China;College of Information Engineering,Southwest University of Science and Technology,Mianyang 621010,Sichuan China;Bazhou Liangjia Agricultural Machinery Manufacturing Co.,Ltd.,Yanqi 841100,Urumqi China)

机构地区：[1]宁波大学科学技术学院,宁波慈溪315300 [2]西南科技大学信息工程学院,四川绵阳621010 [3]巴州良佳农机制造有限公司,新疆焉耆841100

出　　处：《新疆农机化》2024年第4期12-17,共6页Xinjiang Agricultural Mechanization

摘　　要：为实现移栽机的实时监测和统计分析功能,提升移栽作业的整体信息化程度,拓展移栽技术的应用领域,研究了一种适用于移栽机作业质量信息监测的高效识别方法。通过工业相机与边缘计算设备,搭建图像采集与识别平台;利用图像增强技术对移栽环境图像进行扩充;以YOLOv5为基础,通过引入轻量级卷积神经网络MobileNetv2作为特征提取主干,减少模型参数量,采用Relu激活函数降低计算复杂度,缩减卷积计算时间,对改进模型进行量化后训练,以提高运行速度与可部署性,最后采用SORT多目标跟踪算法,实现移栽钵苗质量信息的计数与统计。试验表明:在2ZB-2J高速全自动移栽机正常工作下,平台满足实时性要求,对钵苗识别准确率与统计率较高,可以满足整体系统需求。In order to achieve real-time monitoring and statistical analysis of transplanter,improve the overall informatization level of transplanting operations,and expand the application fields of transplanting technology,an efficient identification method suitable for monitoring the quality information of transplanter operations was studied.Built an image acquisition and recognition platform through industrial cameras and edge computing equipment;Expanded the images of the transplanting environment using image enhancement technology;Based on YOLOv5,a lightweight convolutional neural network MobileNetv2 was introduced as the feature extraction backbone to reduce the number of model parameters.Relu activation function was used to reduce computational complexity and convolutional computation time.The improved model was quantized and trained to improve running speed and deployability.Finally,SORT multi-objective tracking algorithm was used to count and statistically analyze the quality information of transplanted seedlings.The experiment shows that:2ZB-2J high-speed fully automatic transplanter in normal operation,the platform meets real-time requirements,and has high accuracy and statistical rate in identifying bowl seedlings,which can basically meet the overall system requirements.

关 键 词：移栽机 移栽质量 边缘计算 目标跟踪 YOLO 

分 类 号：S24[农业科学—农业电气化与自动化]