plantvillage-full
收藏资源简介:
PlantVillage (full) 是一个经过整理并重新发布的植物病害图像数据集,源自2016年发布的原始PlantVillage数据集。该数据集包含54,304张在受控光照和纯色背景下拍摄的植物叶片图像,涵盖38个不同类别,每个类别对应一种特定的宿主植物物种及其所患的病害或健康状态。数据集提供了结构化元数据层,包括对宿主植物和病害名称的解析、基于物理叶片来源的分组(通过`leaf_id`标识),以及一个确定性的训练集与测试集划分(约80/20比例,实际为43,356张训练图像和10,948张测试图像),确保同一片叶子的所有图像仅出现在训练集或测试集之一,避免了数据泄露。数据以Parquet分片格式打包,便于快速流式访问。每个样本包含图像(统一为256x256像素的RGB JPEG格式)、类别标签(如“Apple___Apple_scab”)、类别索引(0-37)、解析出的宿主植物名称、病害名称、划分标记(“train”或“test”)、叶片ID以及一个指示叶片ID是否源自上游分组元数据的布尔标志。该数据集主要用于图像分类任务,特别是在农业和植物病理学领域的植物疾病自动检测研究,许可证为CC0 1.0(公共领域奉献)。
PlantVillage (full) is a curated and republished plant disease image dataset derived from the original PlantVillage dataset released in 2016. This dataset contains 54,304 plant leaf images captured under controlled lighting and solid-colored backgrounds, covering 38 distinct categories, each corresponding to a specific host plant species and its associated disease or health status. The dataset provides a structured metadata layer, including parsed names of host plants and diseases, grouping based on physical leaf origins (identified via `leaf_id`), and a deterministic train/test split with an approximate 80/20 ratio (specifically 43,356 training images and 10,948 test images). It ensures that all images from the same leaf appear exclusively in either the training set or the test set, preventing data leakage. The data is packaged in Parquet shard format for efficient streaming access. Each sample includes images uniformly formatted as 256x256 pixel RGB JPEGs, category labels (e.g., "Apple___Apple_scab"), category indices (0-37), parsed host plant names, disease names, split markers ("train" or "test"), leaf IDs, and a boolean flag indicating whether the leaf ID originates from upstream grouping metadata. This dataset is primarily used for image classification tasks, especially research on automated plant disease detection in agriculture and plant pathology, and is licensed under CC0 1.0 (Public Domain Dedication).




