不同成熟度番茄果实多模态图像数据集  

A dataset of multimodal images of tomato fruits at different stages of maturity

作　　者：张雨 饶元[1,2,3] 陈文骏 侯文慧 闫胜利 李洋 周传起[2,3,4] 王丰仪 储忧艺 时玉龙 ZHANG Yu;RAO Yuan;CHEN Wenjun;HOU Wenhui;YAN Shengli;LI Yang;ZHOU Chuanqi;WANG Fengyi;CHU Youyi;SHI Yulong(College of Information and Computer Science,Anhui Agricultural University,Hefei 230036,P.R.China;Key Laboratory of Agricultural Sensors,Ministry of Agriculture and Rural Affairs,Hefei 230036,P.R.China;Anhui Provincial Key Laboratory of Smart Agricultural Technology and Equipment,Hefei 230036,P.R.China;College of Engineering,Anhui Agricultural University,Hefei 230036,P.R.China)

机构地区：[1]安徽农业大学信息与计算机学院,合肥230036 [2]农业农村部农业传感器重点实验室,合肥230036 [3]智慧农业技术与装备安徽省重点实验室,合肥230036 [4]安徽农业大学工学院,合肥230036

出　　处：《中国科学数据(中英文网络版)》2025年第1期66-81,共16页China Scientific Data

基　　金：国家自然科学基金(32371993);安徽省重点研究和开发计划面上攻关项目(202104a06020012、202204c06020022、2023n06020057);安徽省高校自然科学研究重大项目(2022AH040125)。

摘　　要：准确识别番茄成熟度并确定最佳收获时间是实现番茄高效采摘并保障采后品质的重要前提。然而,在实际采摘作业场景中,复杂光照导致RGB图像质量下降,限制模型特征提取能力,影响应用效果。此外,现有的番茄串检测和分割数据集难以在番茄异步成熟的生长过程中满足成熟度定向采摘需求。构建基于可见光、深度、近红外的不同成熟度番茄果实多模态图像数据集,能够有效弥补当前研究领域的空白。本数据集采用机械车平台采集和人工辅助采集两种采集方式拍摄视频流数据,通过抽帧降冗、模态对齐、人工筛选等步骤获得4000组多模态图像数据样本,涵盖自然光、人工光、微弱光、钠黄光4种光照条件下的番茄样本图像,包含未熟、半熟、成熟三类成熟度的目标检测和语义分割标注,共计10.08GB。本数据集在YOLOX和Deeplabv3+模型上表现良好,可为番茄智能管理与采收设备的视觉智能系统研发提供基础数据支撑。Accurately identifying tomato ripeness and determining the optimal harvest time are crucial prerequisites for efficient tomato picking and ensuring post-harvest quality.However,in practical harvesting scenarios,complex lighting conditions can degrade the quality of RGB images,limiting the feature extraction capabilities of models and affecting application performance.Moreover,existing datasets for tomato cluster detection and segmentation are insufficient for meeting the requirements of maturitytargeted harvesting during the asynchronous ripening process of tomatoes.Constructing a multimodal image dataset of tomatoes at various stages of ripeness,using visible light,depth,and near-infrared imaging,can effectively fill this research gap.This dataset was collected using both a mechanized platform and manual-assisted methods to capture video stream data.Through frame extraction,redundancy reduction,modality alignment,and manual selection,4,000 multimodal image samples were obtained.The dataset include tomato samples under four lighting conditions:natural light,artificial light,low light,and sodium yellow light.Additionally,it also has annotations for object detection and semantic segmentation at three ripeness stages:unripe,semi-ripe,and ripe,with a total data volume of 10.08 GB.This dataset performs well with YOLOX and DeepLabv3+models and provides fundamental data support for the development of vision-based intelligent systems in tomato’s smart management and harvesting equipment.

关 键 词：番茄 成熟度 多模态 目标检测 语义分割 

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