2020年用于深度学习的多类杂草物种图像数据集

DeepWeeds数据集旨在为基于深度学习的杂草种类分类提供大量杂草图像。该数据集共包含17,509 张红绿蓝图像，这些图像为jpg格式，大小为256×256像素，由定制的地上杂草控制机器人采集，用于自然田间条件下的八种杂草和各种非杂草植物。数据采集过程中无照明控制。每个杂草种类均有1000多张图像，均为训练复杂深度学习模型所需要的图像。该数据集为每个图像提供类标签，尤其有助于研究杂草分类工作。但未提供像素级注释时，不能轻易用于杂草分割和本土化。https://github.com/AlexOlsen/DeepWeeds

The DeepWeeds dataset is designed to provide a large corpus of weed images for deep learning-driven weed species classification. It contains a total of 17,509 RGB images in JPEG format with a resolution of 256×256 pixels, collected by a custom ground-based weed control robot for eight weed species and various non-weed plants under natural field conditions. No lighting control was employed during the data collection procedure. Each weed species is represented by over 1,000 images, which meets the data requirements for training complex deep learning models. The dataset provides class labels for every image, making it particularly conducive to research on weed classification tasks. However, in the absence of pixel-level annotations, it cannot be straightforwardly applied to weed segmentation and localization tasks. https://github.com/AlexOlsen/DeepWeeds