摘要
密集性穴盘苗缺苗穴位的检测,是机械化移栽及后续管理过程中的一项重要工作。为提高缺苗检测的准确率,提出了一种基于深度学习的密集性穴盘苗缺苗穴位的检测方法。检测时,对采集的图像进行穴盘区域自动剪裁,基于YOLOv4卷积网络,提取正常光照、较强光照、苗叶越界、蛭石泛青情况下的缺苗穴位特征进行训练。最后,在测试集上进行试验,结果表明:上述4种条件下缺苗穴位检测的准确率均值为95.2%。此外,与传统图像法相比,该检测方法提高了缺苗穴位在复杂条件下的检测适应性及准确率,能够为温室穴盘育苗模式下苗株个体的生长管理及后续的自动化作业提供依据。
The detection of missing holes in dense plug seedlings is an important work in the process of mechanized transplanting and subsequent management.In order to improve the accuracy of seedling deficiency detection,a detection method of seedling deficiency holes of dense plug seedlings based on deep learning was proposed in this study.The collected images are automatically clipped in the hole disc area.Based on the yolov4 convolution network,the characteristics of missing holes in the case of normal light,strong light,seedling leaves crossing the boundary and vermiculite green are extracted for training.Finally,experiments are carried out on the test set.The results show that the average accuracy of missing seedling hole detection under the above four conditions is 95.2%.In addition,compared with the image method,this detection method improves the detection adaptability and accuracy of missing seedling holes under complex conditions,and can provide a basis for individual growth management and subsequent automatic operation under the greenhouse hole tray seedling mode.
作者
曹丹丹
朱玉桃
王寅初
张鑫宇
卫咏哲
崔永杰
Cao Dandan;Zhu Yutao;Wang Yinchu;Zhang Xinyu;Wei Yongzhe;Cui Yongjie(College of Mechanical and Electronic Engineering,Northwest A&F University,Yangling 712100,China;Shaanxi Key Laboratory of Agricultural Information Perception and Intelligent Service,Yangling 712100,China;Key Laboratory of Agricultural Internet of Things,Ministry of Agriculture and Rural Affairs,Yangling 712100,China)
出处
《农机化研究》
北大核心
2023年第3期210-215,共6页
Journal of Agricultural Mechanization Research
基金
陕西省重点研发计划项目(2019ZDLNY02-04)。
关键词
密集穴盘
缺苗穴位识别
深度学习
图像处理
dense disc
identification of missing seedling hole
deep learning
image processing
