Peach Leaf Damage — Local Field Dataset (Jumilla, Murcia, Spain)

This dataset contains 180 annotated, leaf-level images of peach (Prunus persica) leaves collected under real open-field conditions, intended for image-based classification of foliar damage and for studying domain shift between curated public datasets and operational field imagery. The images were acquired in a 2-ha commercial peach orchard located in Jumilla (Murcia, Spain; 38.438931° N, 1.304664° W), under a semi-arid Mediterranean climate. The orchard was established in 2010 with 'Grocivac 2' (IVIA) cultivar trees grafted onto GF677 rootstock, at a 5.5 m × 3 m spacing. Acquisition took place in two campaigns: an initial set of in-situ images captured in 2024, and a second campaign during the 2025 growing season using a camera module with a 1/1.3" sensor (50 MP, 1.2 µm pixel pitch) and a fixed 24 mm-equivalent f/1.7 lens. Individual leaves were manually segmented and cropped to produce one labelled image per leaf. Each image is assigned to one of four classes reflecting the conditions naturally present at the site during the acquisition period: healthy — 107 images mechanical_stress — 44 images abiotic_stress — 22 images chewing_insect — 7 images Two further classes present in public peach datasets (bacterial_spot and mite_presence) were not observed at the study site during sampling, as the campaign was conducted under late-season conditions with established phytosanitary and biological control management. To preserve ecological validity, no external or simulated samples were introduced. This dataset was used as the target (local) domain for supervised domain-adaptation experiments in the associated publication. The source code is available at https://github.com/adricanovas/peach-leaf-cbam.