Upadhyay Arjun, C Sunil G, Mahecha Maria Villamil, Mettler Joseph, Howatt Kirk, Aderholdt William, Ostlie Michael, Sun Xin
Department of Agricultural and Biosystems Engineering, North Dakota State University, Fargo, ND, USA.
Department of Plant Sciences, North Dakota State University, Fargo, ND 58102, USA.
Data Brief. 2025 Mar 19;60:111486. doi: 10.1016/j.dib.2025.111486. eCollection 2025 Jun.
Effective weed management is crucial for maintaining optimal crop growth and achieve higher yield. Recent advancement in robotic technologies and advanced deep learning (DL) models is shaping the future of robotic weed control systems. However, DL models for weed identification requires substantial amount of data collected in natural field conditions. This article presents red, green, and blue (RGB) datasets for multiple weed species found across different crop production systems. DL models require sophisticated datasets for training the model to achieve high object detection accuracy. To achieve this, a real field dataset was collected under diverse environmental conditions to mimic the natural environment and exhibits the variability in datasets. This aims to improve the accuracy of deep learning models for real time weed identification in precision agriculture. The dataset presented in this article was collected using Canon RGB camera, mounted on the front of remote-controlled robotic platform. This dataset comprises 1120 labelled images presenting five species of weeds and eight different crop species. This resource can be utilized by researchers, educators, and students in developing DL models for weed identification. The dataset can be further enriched by combining it with other relevant weed-crop datasets to create more diverse and robust datasets. This will enhance the capabilities of DL algorithms to be integrated with robotic weed control platforms for precision weed management.
有效的杂草管理对于维持作物的最佳生长和实现高产至关重要。机器人技术和先进深度学习（DL）模型的最新进展正在塑造机器人杂草控制系统的未来。然而，用于杂草识别的DL模型需要在自然田间条件下收集大量数据。本文介绍了在不同作物生产系统中发现的多种杂草物种的红、绿、蓝（RGB）数据集。DL模型需要复杂的数据集来训练模型，以实现高目标检测准确率。为了实现这一点，在不同环境条件下收集了一个真实田间数据集，以模拟自然环境并展示数据集中的变异性。这旨在提高深度学习模型在精准农业中实时杂草识别的准确性。本文介绍的数据集是使用安装在遥控机器人平台前部的佳能RGB相机收集的。该数据集包含1120张标记图像，呈现了五种杂草和八种不同的作物物种。研究人员、教育工作者和学生可以利用该资源开发用于杂草识别的DL模型。通过将该数据集与其他相关的杂草-作物数据集相结合，可以进一步丰富该数据集，以创建更多样化和强大的数据集。这将增强DL算法与机器人杂草控制平台集成以实现精准杂草管理的能力。
