2022年山西农谷番茄小镇温室串番茄果实数据集  

A dataset of greenhouse cluster tomato fruit in the Tomato Town of Jinzhong National Agricultural High-tech Industries Demonstration Zone in 2022

作　　者：宋国柱 石岩 王建 景超 罗改芳 孙胜[2] 王晓丽[3,4] 李一诺 SONG Guozhu;SHI Yan;WANGJian;JING Chao;LUO Gaifang;SUN Sheng;WANG Xiaoli;LI Yinuo(School of Software,Shanxi Agricultural University,Taigu 030801,P.R.China;School of Horticulture,Shanxi Agricultural University,Taigu 030801,P.R.China;Agricultural Information Insitute of CAAS,Beijing 100081,P.R.China;National Nanfan Research Institute(Sanya),Chinese Academy of Agricultural Sciences,Sanya 572024,P.R.China)

机构地区：[1]山西农业大学软件学院,山西太谷030801 [2]山西农业大学园艺学院,山西太谷030801 [3]中国农业科学院农业信息研究所,北京100081 [4]中国农业科学院国家南繁研究院,海南三亚572024

出　　处：《中国科学数据（中英文网络版）》2023年第4期427-434,共8页China Scientific Data

基　　金：山西省基础研究计划项目(202103021224173);山西省科技厅重点研发项目(202102140601015);晋中国家农高区水果番茄智慧标准化技术研究教授、博士工作站资助(JZNGQBSGZZ004)。

摘　　要：串番茄果实的快速、高精度识别,是提升番茄采摘机器人采摘效率、可靠运行的关键技术之一。构建串番茄实时准确检测识别模型,需要有大量的串番茄图像数据进行深度的学习训练。2022年7–8月期间,在山西省晋中市太谷区范村镇格子头村山西农谷番茄小镇对玻璃温室中的串番茄进行数据采集,分别在晴天、阴天,在不同时间段,从不同光位,使用不同型号手机对串番茄进行了多角度、多方位拍摄,经过整理筛查共选出3665张图像,其大小为5.31 GB。使用Label Img工具对选出的图像标注了成熟(mature)、未成熟(raw)、成熟被遮挡(cover)三类标签,并存储为支持yolo格式的TXT文档,其大小为0.8 MB。按训练集:验证集:测试集为8:1:1的比例对所有图像随机分类,使用yolo工具训练并测试串番茄数据集,测试结果的各项性能指标均有不同程度的提升,保证了串番茄数据集的真实有效。本数据集还适用于构建串番茄不同成熟度的卷积神经网络模型,以进一步精准实现串番茄产量预测及成熟度采摘判定等研究。The rapid and high-precision identification of tomato fruit is one of the key technologies to enhancing the picking efficiency and reliable operation of tomato picking robots.Therefore,a substantial volume of tomato image data is essential for in-depth learning and training to build a real-time and accurate identification model of tomato clusters.From July to August 2022,we collected the data of cluster tomato in the glass greenhouse in the Tomato Town of Jinzhong National Agricultural High-tech Industries Demonstration Zone,Getou Village,Fancun Town,Taigu District,Jinzhong City,Shanxi Province.We took pictures of cluster tomato in different angles and directions with different models of mobile phones in different light positionsat different times of the day on sunny days and cloudy days.After sorting and screening,we selected 3,665 images with a size of 5.31 GB.Label Img tool was used to label the selected images with three types of labels:mature,raw and cover,which are stored as TXT documents supporting yolo format,with a size of 0.8MB.We randomly categorized all the images according to an 8:1:1 ratio for the training set,validation set,and test set,respectively.We further adopted yolo tools to train and test the tomato cluster dataset.All performance indicators of the test results have been improved to varying degrees,thereby ensuring the authenticity and effectiveness of the tomato cluster dataset.This dataset can be effectively used for constructing the convolution neural network model of cluster tomato at various stages of maturity,so as to further accurately realize the research on yield prediction and maturity-based picking decisions of cluster tomato.

关 键 词：串番茄 图像 数据集 标注 设施农业 

分 类 号：S641.2[农业科学—蔬菜学] TP391.41[农业科学—园艺学]