JNUSafflower

The JNUSafflower dataset comprises labeled images of safflower variety leaves, cultivated and collected at the Rural Development Administration (RDA) in Jeonju, Republic of Korea. The camera system settings were adjusted following the methodology outlined by Yu et al. (2023) [30]. For imaging, an indoor studio (800 × 800 × 800 mm) was set up to minimize ambient light interference. An 18W white LED (5600K, CN-T96, Plastic, Korea) was used to ensure consistent illumination, preventing data distortion or damage from shadows. A custom-made white background plate was used to account for the color of the safflower leaves. The camera, a Canon EOS D200II (with an EF-S 18-55mm lens, Canon, Japan), features a 24.1-megapixel CMOS sensor. The initial camera settings were ISO 200, a focal length of 35mm, and an exposure time of 1/25. Due to the convex nature of the lens, which can cause image distortion (thicker in the center compared to the edges), we used the camera's built-in distortion correction function. The adjusted settings were ISO 100, a focal length of 18mm, and an exposure time of 1/13. The background was changed to black to contrast with the safflower leaves. Custom 3D panels were designed to accurately represent leaf size. Each panel included crop name tags, four grayscale calibration markers (10mm each), and a QR code to maintain consistent positioning (https://www.group8tech.com/gray-scale-calibration). Images were saved in JPG format with a resolution of 6024 × 4024 pixels. Safflower varieties were grown under controlled environmental conditions, including regulated temperature, irrigation, and fertilizer levels. The experiment took place in an experimental garden containing 43 safflower varieties, each randomly positioned in a greenhouse plot and labeled with a variety code. Six leaf samples were collected for each variety, with sampling conducted before and after the bloom stage throughout the growing season. Leaves were selected based on standardized criteria such as area, weight, and color.

JNUSafflower数据集包含了在韩国全州农村发展管理局（RDA）培养和收集的向日葵品种叶片的标记图像。该数据集的拍摄系统设置遵循了Yu等人（2023）[30]所述的方法论。为了成像，搭建了一个室内摄影棚（800 × 800 × 800毫米），以最小化环境光的干扰。使用了一盏18W的白色LED灯（5600K，CN-T96，塑料，韩国），以确保一致的照明，防止阴影导致的数据失真或损坏。为了考虑向日葵叶片的颜色，使用了定制的白色背景板。相机为佳能EOS D200II（配EF-S 18-55mm镜头，佳能，日本），配备24.1百万像素的CMOS传感器。初始相机设置ISO 200，焦距35毫米，曝光时间1/25。由于镜头的凸形特性（中心比边缘厚），我们使用了相机内置的畸变校正功能。调整后的设置ISO 100，焦距18毫米，曝光时间1/13。背景改为黑色，以与向日葵叶片形成对比。设计了定制的3D面板，以精确地表示叶片大小。每个面板包括作物名称标签、四个灰度校准标记（每个10毫米）和一个QR码，以保持一致的定位（https://www.group8tech.com/gray-scale-calibration）。图像以JPG格式保存，分辨率为6024 × 4024像素。在受控的环境条件下培育了向日葵品种，包括调节温度、灌溉和肥料水平。实验在一个包含43个向日葵品种的实验花园中进行，每个品种随机放置在温室地块中，并贴有品种代码。每个品种收集了六个叶片样本，在整个生长季节的开花前后进行采样。叶片的选择基于标准化的标准，如面积、重量和颜色。