Data for: Deep Learning–Based Identification of Visually Similar Foliar Diseases in Field-Grown Barley

High-throughput assessment of foliar diseases remains one of the main constraints in breeding programs targeting improved resistance. Most existing deep learning datasets focus on single pathogens and controlled imaging conditions, which limits their usefulness in realistic field scenarios where several visually similar diseases often appear together.
To support research on robust multiclass disease segmentation, we assembled a high-resolution image collection of barley flag leaves naturally infected with two economically significant pathogens: brown rust (Puccinia hordei) and ramularia leaf spot (Ramularia collo-cygni). The dataset reflects the complexity of field-grown plants, including mixed infections, heterogeneous symptom expression, senescence-related discolouration, and substantial class imbalance.
Leaves were collected from 62 barley genotypes cultivated in field trials in Estrées Saint Denis (France) and in Irlbach/Paitzkofen (Germany) during developmental stages corresponding to Zadoks 50–69. In total, the dataset contains 336 annotated leaves, which were subsequently converted into 3,632 image patches.

在针对抗病性改良的育种项目中，高通量叶片病害评估仍是主要瓶颈之一。现有深度学习数据集大多聚焦于单一病原菌与可控成像环境，这限制了其在多种视觉相似病害常混合发生的真实田间场景中的应用价值。

为支撑鲁棒多类别病害分割研究，我们构建了一套高分辨率图像数据集，采集对象为自然侵染两种经济重要性病原菌的大麦旗叶：褐锈病（Puccinia hordei）与柱隔孢叶斑病（Ramularia collo-cygni）。该数据集反映了田间种植植株的复杂状态，涵盖混合侵染、症状表达异质性、衰老相关褪色以及严重的类别不平衡问题。

样本叶片采集自法国埃斯特雷-圣但尼（Estrées Saint Denis）与德国伊尔巴赫/派茨科芬（Irlbach/Paitzkofen）的田间试验中种植的62份大麦基因型，采集时期对应扎多克斯（Zadoks）发育分期50~69阶段。本数据集总计包含336张标注叶片，后续被进一步切割为3632个图像切块。