2015年作物和杂草田间图像数据集

作物和杂草田间图像数据集是用于杂草控制的首批公共田间数据集之一。研究者在胡萝卜农场安装自主机器人，通过机器人上的近红外线多光谱相机在现场拍摄收集图像。相机拍摄处采用遮光和人工照明，以避免光照条件变化影响拍摄效果。拍照时选择相机的红色和近红外线通道进行成像（但数据集中的图像保存为红色-近红外线-红色三通道格式）。该数据集总共包含60张分辨率为1296×966像素的png格式原始图像，以及表示植被掩膜的相应二进制图像和定义杂草、作物和土壤背景的像素级注释。注释以三通道图像形式提供，且单独存储在YAML文件中。虽然本数据集体量相对较小，但已可被用于评估用于除草平台的机器人机器学习模型。https://github.com/cwfid/dataset

The Crop and Weed Field Image Dataset is one of the first public field datasets for weed control. Researchers deployed an autonomous robot on a carrot farm, and captured field images using the near-infrared multispectral camera equipped on the robot. Shading and artificial lighting were employed at the camera's shooting site to avoid the impact of varying lighting conditions on image capture. The red and near-infrared channels of the camera were selected for imaging during shooting, yet the images in the dataset are stored in a three-channel format of Red-Near Infrared-Red. This dataset includes a total of 60 original PNG-format images with a resolution of 1296×966 pixels, alongside corresponding binary images for vegetation masks and pixel-level annotations that delineate weeds, crops, and soil background. The annotations are provided as three-channel images and stored separately in YAML files. Although the dataset has a relatively small scale, it can already be utilized to evaluate robotic machine learning models for weed control platforms. The dataset is accessible at https://github.com/cwfid/dataset