WheatPhenology: A Multi-Stage Field Image Dataset of Wheat Growth

The image database name WheatPhenology is the field image repository directly designed to support created inquiries on crop phenology monitoring, agricultural computer vision, and precision agricultural practices. This repository contains high-resolution images of wheat crops collected at various phenological stages and, thus, can be utilized to train and test machine learning and deep-learning models to identify crop phenological stages. The farming fields within the area of 60-kilometre around the city of Sehore in Madhya Pradesh, India, a region known to have a high wheat production and a heterogeneous field environment were systematically sampled. This spatial heterogeneity makes the dataset able to reflect the differences in soil type, irrigation schedule, crop plant density, luminance and background, which reflect the real farming environments.

The data has been painstakingly recorded in the form of wheat growth covering the key stages in development of the crop, including early vegetative growth, tillering, elongation of stems, booting, heading, flowering, grain filling, and maturity. The dataset, where the phenological phases are presented in the form of a visual representation, can be considered an invaluable source of training to algorithms aimed at identifying crop-stage classification or yield forecasting, growth monitoring, as well as developing intelligent agricultural decision-support systems. The photographs were taken in a real field setting using regular digital cameras and cell phone cameras, thus, guaranteeing the realistical change of illumination, perspective, and position of plants. This kind of acquisition strategy increases the extrapolation of models trained using the data to practical agricultural conditions.

The dataset used to achieve robustness and application to the computer vision applications is inclusive of the images provided by different farms, different angles of the camera, and at various points in time during the wheat growing season. The use of the collection methodology focuses on the natural variability in crop structure, canopy formation and environmental background. Therefore, WheatPhenology provides an all-inclusive depiction of wheat growth in working farms. The dataset can be used to study the plant phenotyping, crop monitoring, phenological stage identification, and agricultural management systems based on AI as a benchmark. The dataset by combining spatial heterogeneity with the multi-phased growth representation can play a significant role in the development of automated crop monitoring systems and agricultural research based on data.

图像数据集WheatPhenology是专为支撑作物物候监测、农业计算机视觉与精准农业实践等领域的相关研究而设计的实地图像库。该库收录了不同物候期采集的小麦作物高分辨率图像，可用于训练和测试用于识别作物物候期的机器学习与深度学习模型。
研究团队对印度中央邦塞霍尔市周边60公里范围内的农田开展了系统采样——该区域不仅小麦产量颇高，且田间环境异质性显著。这种空间异质性使得该数据集能够反映土壤类型、灌溉方案、作物种植密度、光照条件与背景环境的差异，从而贴合真实农田场景。
该数据集细致记录了小麦生长周期内的关键发育阶段，涵盖营养生长早期、分蘖期、茎秆伸长期、孕穗期、抽穗期、开花期、灌浆期与成熟期。该数据集以可视化形式展示作物物候阶段，可作为训练面向作物阶段分类、产量预测、生长监测以及智能农业决策支持系统开发算法的宝贵资源。
本次图像采集采用普通数码相机与手机相机在真实农田环境中完成，因此能够还原光照条件、拍摄视角与植株空间位置的真实变化。此种采集策略可提升基于该数据集训练的模型在实际农业场景中的泛化性能。
为提升模型鲁棒性并适配计算机视觉应用，该数据集涵盖了不同农场、不同拍摄角度以及小麦生长季不同时间点采集的图像。本次采集方法聚焦于保留作物结构、冠层形态与环境背景的自然变异性。因此，WheatPhenology数据集可全面展现实际耕作农田中小麦生长的全周期图景。该数据集可作为植物表型分析、作物监测、物候阶段识别以及基于人工智能的农业管理系统研究的基准数据集。该数据集将空间异质性与多阶段生长表征相结合，可为自动化作物监测系统研发以及数据驱动的农业研究发挥重要作用。