Image Dataset of Grapevine Varieties for AI-based Cultivar Identification

This dataset comprises multi-spectral images of seven grapevine varieties, captured under field conditions using an autonomous ground vehicle (AGV) equipped with a custom-built imaging rig. The rig integrates three synchronized cameras mounted at varying heights and angles to capture comprehensive visual data from both sunlit and shaded sides of the vine rows. Canon PowerShot SX260 HS (1.80 m) Canon PowerShot S100 (0.35 m) MAPIR Survey3W RGB (1.00 m, horizontal view) All cameras were configured to capture images at intervals of approximately 0.5 to 1 second, ensuring synchronized data acquisition across different spectral bands and perspectives. The primary objective of this dataset is to facilitate the development and evaluation of computer vision and machine learning algorithms for automatic grapevine cultivar identification. Each grapevine variety is represented by hundreds of images, systematically organized by camera type, illumination condition (sunlit/shaded), and acquisition date. To optimize storage efficiency and processing speed, images are provided in JPEG format; RAW files are not included. Canon cameras were controlled using CHDK (Canon Hack Development Kit) and programmed via the HCKY/XXXUAV scripts to enable interval shooting, RAW support, and manual configuration. The MAPIR camera were configured for automatic capture at 0.5–1 second intervals and saved in JPG format. Images were collected from both sides of each vine row to capture sunlit and shaded leaf conditions (referred to as "ILUMINADO/SOMBREADO" or "ESTE/OESTE"). No RAW images are included to reduce dataset size and optimize acquisition speed. The dataset is divided into two image sets (Lot 1 and Lot 2), taken in October 2024. Image Summary by Lot and Variety Lot 1: 6,704 images Lot 2: 6,337 images Total: 13,041 images Breakdown by variety: Prieto Picudo: 2,565 images Albillo Real: 1,445 images Verdejo: 2,004 images Mencía: 1,538 images Garnacha: 1,534 images Albillo Mayor: 1,378 images Tempranillo: 877 images本数据集涵盖7个葡萄品种的多光谱图像，所有图像均采集于野外环境，由搭载定制化成像装置的自动地面载具（AGV）完成拍摄。该装置集成三台同步相机，各相机以差异化高度与角度安装，可全面采集葡萄藤列阳面与阴面的视觉数据。 佳能PowerShot SX260 HS相机（安装高度1.80米） 佳能PowerShot S100相机（安装高度0.35米） MAPIR Survey3W RGB相机（安装高度1.00米，水平视角） 所有相机均配置为以0.5至1秒的近似间隔拍摄图像，以保障不同光谱波段与拍摄视角下的数据采集同步。 本数据集的核心目标是辅助用于自动葡萄品种识别的计算机视觉与机器学习算法的研发与评估。 每个葡萄品种均对应数百张图像，并按相机类型、光照条件（阳面/阴面）及采集日期进行系统化归档整理。为优化存储效率与处理速度，图像以JPEG格式提供，不包含RAW原始文件。 佳能相机通过CHDK（佳能黑客开发套件，Canon Hack Development Kit）进行控制，并通过HCKY/XXXUAV脚本编程，以实现间隔拍摄、RAW文件支持及手动参数配置。MAPIR相机设置为以0.5至1秒的间隔自动拍摄，图像保存为JPG格式。 图像采集自每一条葡萄藤列的两侧，以获取阳面与阴面的叶片图像（对应标注为"ILUMINADO/SOMBREADO"或"ESTE/OESTE"）。本数据集未包含RAW原始文件，以缩减整体存储体积并优化采集效率。数据集分为两个图像批次（批次1与批次2），均采集于2024年10月。 按批次与品种的图像统计如下： 批次1：6704张图像 批次2：6337张图像 总计：13041张图像 按品种的图像分布： 普列托·皮库多（Prieto Picudo）：2565张图像 阿尔维略·雷亚尔（Albillo Real）：1445张图像 维德霍（Verdejo）：2004张图像 门西亚（Mencía）：1538张图像 加尔纳恰（Garnacha）：1534张图像 阿尔维略·马约尔（Albillo Mayor）：1378张图像 丹魄（Tempranillo）：877张图像