摘要
设施番茄疏花疏果工作多依赖于人工,针对人工逐一判断、工作量大、移动设备实时性要求高等问题,通过卷积神经网络进行番茄部分生长参数识别,并结合农事操作经验进行应用讨论。在卷积神经网络SSD算法中引入轻量化模块MobileNetV3,提出了基于SSD-MobileNetV3模型的番茄花和果实的识别分析方法。与传统方法相比,一定程度上克服了重叠及遮挡、光照条件变化、亮度不均等干扰因素的影响。在串开花数、串结果数等生长参数识别基础上,结合疏花疏果工作进行讨论。结果表明,试验模型对于设施环境下常见干扰因素具有良好的实时性和鲁棒性,与SSD算法相比,花果平均识别率为92.57%,提高了7.9%,识别速度为0.079 s,提升了约4倍,识别率和识别速度明显提高,计算参数减少,基本满足应用要求。
Aiming at the problems of facility tomato flower thinning and fruit thinning relying on manual judgment one by one,the workload is large,and the real-time requirements of mobile equipment are high,here we use convolutional neural network to identify some tomato growth parameters,and discusses the application based on agricultural operation experience.The lightweight module MobileNetV3 is introduced into the convolutional neural network SSD algorithm,and a method for identifying and analyzing tomato flowers and fruits based on the SSD-MobileNetV3 model is proposed.Compared with the traditional method,it overcomes the influence of interference factors such as overlap and occlusion,changes in lighting conditions,and uneven brightness to a certain extent.On the basis of the identification of growth parameters such as the number of bunches and the number of bunches,the discussion was carried out in conjunction with the work of flower thinning and fruit thinning.The results show that the experimental model has good real-time performance and robustness for common interference factors in the facility environment.Compared with the SSD algorithm,the average recognition rate of flowers and fruits is 92.57%,which is an increase of 7.9%,and the recognition speed is 0.079 s,which is an increase of approximately 4 times,the recognition rate and recognition speed are significantly improved,and the calculation parameters are reduced,which basically meets the application requirements.
作者
陈新
伍萍辉
祖绍颖
徐丹
张云鹤
董静
CHEN Xin;WU Pinghui;ZU Shaoying;XU Dan;ZHANG Yunhe;DONG Jing(Beijing Agricultural Intelligent Equipment Technology Research Center,Beijing 100097,China;School of Electronic Information Engineering,Hebei University of Technology,Tianjin 300401,China;Tianjin Agricultural Mechanization Technology Test Service Center,Tianjin 301616,China;Beijing Jixing Agriculture Co.LTD,Beijing 101500,China)
出处
《中国瓜菜》
CAS
北大核心
2021年第9期38-44,共7页
China Cucurbits And Vegetables
基金
北京市百千万人才工程项目(2018A33)
重庆市渝北区科技计划项目(2020-30)
北京市农林科学院创新能力建设专项(KJCX20200533)。
关键词
番茄
设施栽培
生长监测
疏花疏果
轻量化神经网络
图像识别
Tomato
Protected Cultivation
Growth monitoring
Sparsely flowered and fruited
Lightweight neural network
Image recognition
