Data for: Deep Learning–Based Identification of Visually Similar Foliar Diseases in Field-Grown Barley
Description
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.
Files
Steps to reproduce
To preserve leaf morphology, collected flag leaves were flattened using adhesive sheets on A4 paper. High-resolution images were obtained by scanning at 800 DPI using an Epson Perfection V600 flatbed scanner under standardised lighting conditions, producing TIFF-format images suitable for downstream analysis. Pixel-level annotations were manually created for brown rust and ramularia lesions, with oversight from expert barley pathologists to ensure accuracy. To facilitate training of deep learning models, the full-resolution leaf images were subsequently subdivided into smaller patches.
Institutions
- Technical University of MunichBavaria, Munich